Abstracts Track 2023

Nr:	24
Title:	MODEC: An Unsupervised Clustering Method Integrating Omics Data for Identifying Cancer Subtypes
Authors:	Yanting Zhang and Hisanori Kiryu
Abstract:	The identification of cancer subtypes can help researchers understand hidden genomic mechanisms, enhance diagnostic accuracy and improve clinical treatments. With the development of high-throughput techniques, researchers can access large amounts of data from multiple sources. Because of the high dimensionality and complexity of multi-omics and clinical data, research into the integration of multi-omics data is needed, and developing effective tools for such purposes remains a challenge for researchers. In this work, we proposed an entirely unsupervised clustering method without harnessing any prior knowledge (MODEC). We used manifold optimization and deep-learning techniques to integrate multi-omics data for identifying cancer subtypes and analyzing significant clinical variables. Since there is nonlinearity in the gene-level datasets, we used manifold optimization methodology to extract essential information from the original omics data to obtain a low-dimensional latent subspace. Then, MODEC uses a deep learning-based clustering module to iteratively define cluster centroids and assign cluster labels to each sample by minimizing the Kullback–Leibler divergence loss. MODEC was applied to six public cancer datasets from The Cancer Genome Atlas database and outperformed eight competing methods in terms of the accuracy and reliability of the subtyping results. MODEC was extremely competitive in the identification of survival patterns and significant clinical features, which could help doctors monitor disease progression and provide more suitable treatment strategies. Three types of gene-level omics datasets from the TCGA portal (https://www.cancer.gov/tcga) were used: gene expression (RNA), miRNA, and DNA methylation data. TCGA subtype datasets were downloaded using the TCGABiolinks R package, whereas clinical datasets were downloaded from the TCGA portal (https://www.cancer.gov/tcga). TCGA molecular subtypes are reported by the TCGA Research Network. Several keywords are used to identify clinical parameters, such as survival and cancer staging. We only kept parameters with less than 10% missing values. Experiments on six cancer datasets from the TCGA database showed that MODEC outperformed eight competing methods based on four external indices, survival analysis, and enrichment analysis. Survival and enrichment analyses verified the practical significance of this work, particularly in terms of assisting doctors to obtain a clinical diagnosis. Compared with the eight benchmark algorithms, the proposed method was capable of identifying significant survival differences and clinical patterns using MODEC subtypes. Additionally, MODEC was extremely robust to the rank choices. In the future, we hope to further improve this method to handle partial datasets and speed up the algorithm's execution.

Nr:	38
Title:	Wearable Device for Home Sleep Apnea Screening
Authors:	Fernanda Irrera, Alessandro Manoni, Daniela Pellegrino, Alessandro Gumiero, Luigi Della Torre and Paolo Palange
Abstract:	Sleep-related breathing disorders (SRBD) are characterized by recurrent episodes of pause or reduction of breathing during sleep that last at least 10 seconds. Severity is evaluated by the Apnea-Hypopnea Index (AHI), defined as the number of events divided by the monitoring time (in hours). Obstructive apneas (OSA) consist in a physical obstruction of the upper airways; central ones (CSA) are due to a transient loss of neural output to the respiratory muscles. People who suffer of SRBD, complain of cognitive deficit and daytime sleep attacks and are more exposed to systemic arterial hypertension, myocardial ischemia and cerebrovascular diseases, with even sudden cardiac death if not treated. It is known that the therapeutic strategies for OSA and CSA are different, so screening apneas and distinguishing the type is fundamental . In spite of this, unfortunately, the diagnosis of SRBD is still a challenge. Indeed, the gold standard to diagnose SRBD is the polysomnography (PSG), which consists in recording signals during the night to recognize sleep stages, monitor cardiac activity and respiratory function. Unfortunately, PSG is very cumbersome, expensive and scarcely available, requires hospitalization and long waiting times and only a small percentage of patients achieves a definitive diagnosis and is followed-up. The huge demand for PSG supports the use of alternative approaches, possibly providing unobtrusive and long-term evaluation. It is evident that the possibility of using wearable devices for the domestic screening and subsequent follow-up of SRBD would be a breakthrough. We developed a wearable wireless device for home monitoring of SRBD, which can be comfortably positioned on the nasal septum during the night. It integrates PPG sensor to study the respiration and an accelerometer to monitor movements. Positioning the device on the nasal septum allows for maximum PPG sensitivity to airflow modulation and for an excellent sensitivity of the accelerometer to thoraco-abdominal movements. A method based on the fusion of PPG and accelerometer signals permits to distinguish OSA from CSA. The device records and transmits data in streaming mode via Bluetooth to an external elaborator. Algorithms have been developed for: a) the apnea detection, using only the PPG signal; b) the apnea type classification, using the fusion of PPG and ACC signals; c) the body position identification, using only the ACC. Performances of our system were tested on a hospitalized patient wearing our integrated system and contemporarily performing the PSG along the whole night. The patient experienced 545 apnea episodes. We achieved a sensitivity of 89% and a precision of 93% in the apnea detection, with a practically coincident AHI index, compared to the gold standard. We obtained such excellent detection performance thanks to the utmost PPG signal modulation on the nose. On the basis of results, our device seems potentially able to: 1) detect any type of apnea and their timing using the PPG signal; 2) detect the chest respiratory efforts thus paving the way for a valid method to distinguish the obstructive and central apnea types, by fusion of the PPG and the ACC signals; 3) grade the SRBD severity calculating the AHI; 4) identify the patient body position during the sleep, using the ACC signal. All this makes our device potentially useful for pervasive home screening SRBD shortcutting low queues for PSG.

Nr:	40
Title:	Hypoglycemia detection and prediction using a wearable device
Authors:	Giordana Di Berardino, Enrico Fornito, Federico Boscari, Angelo Avogaro and Fernanda Irrera
Abstract:	Introduction: Diabetes is a metabolic disorder characterized by high blood glucose that affects, today, approximately half a billion persons worldwide, with an estimated year cost of around 10% of the global health expenditure. Type 1 diabetes (T1D) is due to insulin deficiency: it is treated pharmacologically by insulin administration, which in general cannot replicate precisely the physiological secretion, and easily leads to dangerous glucose oscillations. Hypoglycemia is a significant threat for people with T1D, as it influences physical and mental abilities, and may lead to life-threatening heart arrhythmias. T1D patients suffer, in average, thousands of symptomatic hypoglycemia episodes in their disease lifetime, in particularly during night; this can be dangerous because hypoglycemia symptoms may be blurred by sleep, resulting in coma and even death. For these reasons, different approaches have been recently proposed to identify characteristic features in electrocardiogram (ECG) or electroencephalogram (EEG) before the onset of hypoglycemia episodes. Our proposal: This work proposes an innovative wearable device recording simultaneously the ECG and the EEG alpha waves for hypoglycemia prediction. First, we identified the minimum set of parameters that can be extracted from ECG and EEG in time (amplitude) and in frequency (power spectral density-PSD-) domains that alter hundreds of seconds before the onset of a hypoglycemia episode. Then, we designed a compact, comfortable, easy-to-use, wearable device able to record the two biopotentials synchronously. The critical parameters derived from ECG are calculated from the heart rate variability (HRV): the RR tracts, the RR standard deviation normal to normal (SDNN), the root mean square of RR standard deviation (RMSSD), the low frequency/high-frequency ratio (LF: HF). The critical parameters calculated from the EEG alpha waves are: the frequency centroid (CF) and the spectral rotational radius derived from spectral moments (G0 and higher). The hard system consists of two integrated boards (each mounting an electrostatic sensor, a microcontroller, a battery, a microSD) fixed on the chest, as displayed in Fig.1: one sensor acquires ECG with two electrodes in RA-LL positions, the other sensor acquires the EEG with two electrodes on the occipital sites O1 and O2. The advantages of our system are: 1) it is wearable and comfortable, with just one chest band and four electrodes; 2) the battery life is very long (days) thanks to the sensor very low power consumption; 3) data are stored locally on a microSD; 4) its cost is extremely low thanks to the integrated electronics technology. Preliminary verification of correct operation has been performed by comparison with a gold standard (GS: the Cyton Daisy scientifically validated research tool). Six control subjects have been tested for short times or overnight: they were asked to put on this wearable device and contemporarily the GS, with electrodes of the two systems as close as possible. Results: In Fig. 2 the HRV trends extracted from ECG are compared in time and frequency domain as well as in Fig.3 for EEG alpha waves. The Table reports values of the key parameters calculated from ECG and EEG of one tested subject averaged on few hours recording. The comparison between our system and the GS is excellent, thus candidating our wearable device as an ideal tool for hypoglycemia prediction. Pictures of the two sensors are shown in Fig.4.

Nr:	54
Title:	Insights from Analyses of Low Complexity Regions with Canonical Methods for Protein Sequence Comparison
Authors:	Patryk P. Jarnot, Joanna Ziemska-Legiecka, Marcin Grynberg and Aleksandra Gruca
Abstract:	Low Complexity Regions (LCRs) are fragments of protein sequences consisting of only a few types of residues. Although they look simple, they occur frequently in proteins and therefore should be considered as their important part. In the past, scientists thought that LCRs are irrelevant to the biological roles of proteins. However, these fragments are currently the subject of numerous research which show their importance in protein functions and structures. Since LCRs have been a secondary concern for a long time, we hypothesised that existing methods for similarity analysis of protein sequences were developing to detect similarity between High Complexity Regions (HCRs) only and are insufficient for LCR comparison. In this study we analysed the performance of protein sequence comparison methods for two distinct datasets containing LCRs and HCRs. We extracted LCRs from proteins using SEG tool and used them to create the first dataset. Next we joined all remaining parts of sequences and put them into HCR dataset. For both datasets we run BLAST, HHblits and CD-HIT which are widely used methods to find similar protein sequences. For HCR dataset we run them using the default parameter sets but for LCRs we have optimised parameters to find more results while maintaining reasonable quality. We investigated results by calculating overlap between methods and analysing selected examples from each method. To calculate overlap we connected reported similarities into pairs and drew Venn diagrams. For BLAST and HHblits we connected aligned sequences. For CD-HIT we combined all possible pairs of sequences in each cluster. Most of the protein families were assigned based on HCRs thus we also removed paris in which both sequences belong to the same family. We used pairs with different families assigned in order to select non obvious cases for detailed analysis of selected examples. Finally, we provide these features of each method that are suboptimal for LCR comparison even with optimised parameters. (Co-financed by the EU grant POWR.03.02.00-00-I029/17).

Nr:	57
Title:	Growth Mechanics: General Principles of Optimal Cellular Resource Allocation in Balanced Growth
Authors:	Martin Lercher and Hugo Dourado
Abstract:	The physiology of biological cells evolved under physical and chemical constraints such as mass conservation, nonlinear reaction kinetics, and limits on cell density. For unicellular organisms, the fitness that governs this evolution is mainly determined by the balanced cellular growth rate. Mathematical models are an important tool to understand and predict the complex behavior of biological cells that arose from this evolution. However, prevalent modeling frameworks, which rely on simplified linear optimizations, cannot describe this full complexity. The next generation of more realistic cell models will depend on an efficient mathematical formulation for the corresponding nonlinear optimization problem that facilitates the analytical study and numerical simulation of large models. A previous theoretical study introduced Growth Balance Analysis (GBA) as a general framework to model nonlinear mathematical descriptions of growing cells, presenting analytical conditions for optimal balanced growth in the special case that the active reactions are known. Here, we develop Growth Mechanics (GM) as a more general, succinct, and powerful analytical description of the growth optimization of GBA models, which we formulate in terms of a minimal number of dimensionless variables. Growth Mechanics uses Karush-Kuhn-Tucker (KKT) conditions in a Lagrangian formalism. It identifies fundamental principles of optimal resource allocation in GBA models of any size and complexity, including the analytical conditions that determine the set of active reactions at optimal growth. We identify from first principles the economic values of biochemical reactions, expressed as marginal changes in cellular growth rate; these economic values can be related to the costs and benefits of proteome allocation into the reactions’ catalysts. Our formulation also generalizes the concepts of Metabolic Control Analysis to models of growing cells. Growth Mechanics unifies and extends previous approaches of cellular modeling and analysis, putting forward a program to analyze cellular growth through the stationarity conditions of a Lagrangian function. It thereby provides a general theoretical toolbox for the study of fundamental mathematical properties of balanced cellular growth.

Nr:	59
Title:	Hybrid Modelling to Solve Optimal Concentrations of Metabolites and Enzymes in Constraint-Based Modelling
Authors:	Sabine Peres
Abstract:	The Constraint-based modelling is a widely used approach to analyze genotype-phenotype relationships. The main key concepts are stoichiometric analysis such as flux balance analysis (FBA), Resource Balance Analysis (RBA) or elementary flux mode (EFM) analysis. While FBA identifies optimal flux distribution with respect to a given objective, the EFM characterizes the totality of the available solution space in terms of minimal pathways but their number leads to a combinatorial explosion for large networks. The RBA predicts for a specific environment, the set of possible cell configurations compatible with the available resources and extends very significantly the predictive power of the FBA. However, when stoichiometric and kinetic constraints are considered together, the set of possible flux configurations does not generally define a convex set since the kinetic function are not linear. The problem resolution has thus multiple local maxima. Recent works showed that the optimal solution of constraint enzyme allocation problems with general kinetics is an EFM. Based in this recent outcome, we write the resource allocation constraint on kinetic optimization problem into a geometric problem in an EFM, i.e. a convex optimal problem easily solved. Thus to predict optimal flux modes, we compute constraints EFMs with our tool ASPefm based on Answer Set Programming to save time and space computation. ASPefm allows the integration of Boolean and linear constraints such as thermodynamic, environment, transcriptomic regulatory rules, and resource operating cost (that identify the most efficient EFMs for converting substrate into biomass) using the solver ClingoLP which combined logic programming and linear programming. The convex optimization problem is then resolved on each constraint EFMs which provides for this mode, the optimal repartition of resources among enzymes and the associated metabolite concentrations. We applied our method to the central carbon metabolism of E. coli, with a detailed model of the respiration chains, ATPase (including explicitly the proton motive force). The optimal flux mode is the overflow of acetate which is in agreement with known experimental results. This approach allowed us to explore whether certain experimental properties observed on E. coli are consistent and consequences of an optimal repartition of bacterial resources. Our method is very promising in synthetic biology and increased the ability to efficiently design biological systems.

Nr:	60
Title:	Determination of Polarization Correlations of Gamma-Rays from Positronium Annihilation and Implications to Positron Emission Tomography
Authors:	Ana Marija Kožuljevic, Tomislav Bokulić, Damir Bosnar, Ivica Frišcic, Zdenka Kuncic, Siddharth Parashari, Luka Pavelić, Petar Žugec and Mihael Makek
Abstract:	During the positron emission tomography (PET) scan, decaying radio-pharmaceuticals emit a positron, which undergoes annihilation with an electron from the surrounding tissue. In this process, two photons are emitted back-to-back, with 511 keV energies, and are subsequently detected by the scanner. These two photons also have orthogonal polarizations, a fundamental property which is not yet utilized in conventional PET devices. This polarization correlation offers an opportunity to improve the reconstructed image quality by reducing the random background noise, which lacks this property. To test this possibility, a novel type of a PET scanner, the Quantum-PET Demonstrator, has been developed and commissioned. The device consists of an aluminium ring with four mounted detector modules, which rotates around the source of annihilation. The detector modules consist of four 8 x 8 crystal (GAGG:Ce and/or LYSO:Ce) matrices, with either 2.2 mm or 3.2 mm pitch. Hence, the Demonstrator has 1024 pixels divided into 16 trans-axial rings. The identically pitched modules, mounted opposite of one another on the ring, can determine and reconstruct the polarization correlations of the emitted annihilation quanta by measuring the azimuthal angles of the Compton scattered photons in the modules. We will present the polarimetric performance of the Quantum-PET Demonstrator tested in a laboratory with a Na-22 source. Data acquisition and processing is performed with TOFPET2 ASIC readout system and analysed with different event selection criteria. Potential use of the Quantum-PET Demonstrator in clinical setting will also be discussed.

Nr:	5
Title:	Particle Tolerance of Micropumps for Automated Cell Culture
Authors:	Agnes Bußmann, Nivedha Surendran and Martin Richter
Abstract:	Microfluidic systems enable impressive advances in the field of cell culture. Microscale experiments often require active fluid transport, which is up to date commonly achieved with large pumps. Improvement is possible with micropumps that enable energy-efficient liquid transport with a small, disposable device. However, the pump needs to be robust and able to transport liquids that contain particles, cells or debris. In this work, we experimentally investigate the particle tolerance of piezoelectric micropumps for cell culture applications. INTRODUCTION The transport of liquid samples and reagents is a ubiquitous task in microfluidic biomedical experiments. Up to date, bulky external pumps that require connectors and large tubing, often outside the incubator, are common. The integration of micropumps into small and cost-efficient disposables is promising, as it enables many parallel experiments, and automated fluid transport in the incubator without manual handling [1]. The use in biomedical experiments brings along challenging conditions: The incubator is humid, requiring hermetic sealing, space is limited, and the transported liquids are not clean deionized water and can contain small particles, cells or debris. Hence, it is crucial to investigate the robustness of the microfluidic actuator for use in automated cell culture setups. METHODS The presented work examines the particle tolerance of stainless steel micro diaphragm pumps that displace fluid with a piezoelectric bending actuators [2]. The devices consist of a pump body including passive check valves as well as an actuator diaphragm with a glued on piezoelectric disc actuator. An alternating high voltage signal causes the piezoelectric actuator to contract and expand and induces, in combination with the passive valves, a directed fluid flow. Particles can harm the pump by accumulation between the valve and its seat (causing leakage), in the chamber, which can limit the actuator stroke, or in the fluid path where they block the system. The impact of particle transport is evaluated with 1 µm polystyrene particles pumped for 10 min. The fluidic performance is evaluated before and after particle transport. RESULTS The actuator stroke remains unchanged, though the transport of particles causes a decrease in fluidic performance. The flow without backpressure stays stable, but the capability to pump against pressure is slightly de-creased. This is likely caused by agglomeration of particles in the area of the valve seat. Microscopic evaluations un-derline this theory. CONCLUSION The ability to transport particles, cells, or debris is crucial for robust fluid transport in microfluidic systems. First evaluations show that polystyrene particles influence the pump’s performance only slightly. No pump failed during the experiment and a cleaning procedure enabled to recover original performance. In further experiments, the tolerance towards fibers has to be evaluated. Furthermore, technical adaptations such as surface treatments, geometrical changes and an adaptation of the actuation principle can further improve the pump's robustness. REFERENCES [1] M.-H. Wu et al., Lab on a Chip, vol. 10, no. 8, pp. 939–956, 2010, doi: 10.1039/b921695b. [2] A. Bußmann et al., Sensors and Actuators A: Physical, p. 112649, 2021, doi: 10.1016/j.sna.2021.112649.

Nr:	10
Title:	Using Prior Knowledge to Convert Raw Data into Images and Classify Them: The DPIX Algorithm - Application in a Biomedical Context
Authors:	Nathan Foulquier, Alain Saraux, Jacques Olivier Pers and Pascal Redou
Abstract:	We present a new classification approach that is particularly well suited to biomedical datasets. We named this approach Data Pixelisation (DPIX). DPIX is designed to use prior knowledge during the classification process by calculating a distance between the variables used to describe the observation in the dataset. These distances are computed from the scientific literature using natural language processing approaches (NLP) and are then used to create an image structure in which each variable is converted to a pixel and the nearby pixels represent the near variables. To learn the optimal image structure we designed an optimisation algorithm, which we named Guided Mutation Algorithm (GMA), also presented in this article. The generated images are then used to train a Convolutional Neural Network (CNN) to solve the classification problem. We finally present applications of this technique through three examples and discuss the use of prior knowledge for classification in a biomedical context.

Nr:	12
Title:	Combining Hand-Engineered and Deep-Learning Features Using XGBoost for Neurofibroma Classification in Whole-Body MRI Images
Authors:	Andy Xu, Guibo Luo and Wenli Cai
Abstract:	Purpose: Neurofibromas are tumors on the nerves resulting from the neurofibromatosis genetic condition. They can form in any body part and take on various shapes and sizes. Tumors are either discrete or plexiform where discrete tumors only take up one nerve fascicle while plexiform ones take up multiple nerve fascicles and are much more likely to become malignant. After taking a whole-body MRI (WBMRI), radiologists need to generate a summary report for each patient, recording the volume, body part category, and type of each neurofibroma. The process is very time-consuming, especially when the patient has a lot of tumors, as it takes up to a week for radiologists to finalize the chart. This study aims to enable a fully automated, end-to-end tool to help radiologists automatically generate a summary chart for patients from a WBMRI. Materials and Methods: A dataset of 150 WBMRI scans of patients with neurofibromatosis was collected with 583 discrete tumors and 470 plexiform tumors. The WBMRIs were segmented by hand and the segmentation masks were labeled with tumor types and body part categories. With this dataset, a deep-learning model like UNet can be trained to segment the neurofibromas from a WBMRI. Each tumor is then separated from others and its volume can be automatically measured. The cross-section images of tumors are used to train a Convolutional Neural Network (CNN) for extracting deep-learning features. These features are combined with hand-engineered features such as the volume, center location, bounding box, surface-volume ratio, etc. to train an XGBoost based classifier to predict the body part category and type of each tumor. Results The classification models achieved an accuracy of 92.4% for predicting the body part category out of 12 options and 0.93 AUC for predicting the type as discrete or plexiform, which outperforms the models with only hand-engineered or deep-learning models by a large margin. Using these models, generating a summary chart is much more efficient as it can be done automatically and almost instantly. Conclusion As current research is mostly focused on the segmentation of tumors, this study tries to close the gap for an end-to-end automation pipeline by using the WBMRI and its neurofibroma segmentation masks to automatically generate a summary report with essential details including the body part category and type of each tumor. Clinical Relevance The classification models allow a summary chart to be quickly generated from a WBMRI, significantly improving the speed at which radiologists can interpret the scan with high accuracy.

Nr:	14
Title:	Pedestrian Traffic Light Detection and Recognition Systems for Blind People
Authors:	Eduardo Pinos, Francisco Coronel-Berrezueta and Maria Cordero-Mendieta
Abstract:	This article deals with a very important topic since we have considered that one of the greatest challenges faced by a person with visual impairment, whether blind or low vision, who tries to develop independently, has a palpable difficulty when moving from one place to another as he/she has to cross the streets of a city. Therefore, we present several designs of personal electronic assistive devices for people who suffer from blindness proposed by some authors with innovative solutions that allow to locate pedestrian traffic lights and their status (passing/not passing) in an auditory way, enabling the blind to better manage and gain independence in their environment. Most of the authors have made use of robust systems such as deep neural networks with different configurations in such a way that the solutions found provide the user with an accuracy of more than 90%, avoiding errors that could endanger the life of the blind pedestrian.

Nr:	15
Title:	BiomiX: A User-Friendly Bioinformatic Tool for Automatized Multiomics Data Analysis and Integration
Authors:	Cristian Iperi, Álvaro Fernández-Ochoa, Jacques Olivier Pers, Divi Cornec, Anne Bordron and Christophe Jamin
Abstract:	1. Objective The usage of high-throughput technology in health and biological sciences boosted the amount of information obtainable from samples, ensuring highly robust disease diagnosis and consistent research approaches. The increased dependency on these technologies revealed how data analysis represents the bottleneck step both in time and in skilled bioinformatics users. The BiomiX tool offers an efficient and fast pipeline to analyze -omics data individually and integrate multi-omics data from the same patients. 2. Methods BiomiX has been developed from the European PRECISESADS database (support from the Innovative Medicines Initiative Joint Undertaking under the Grant Agreement Number 115565)1, including overlapping data of whole blood and sorted immune cells transcriptomic, serum and urine metabolomic, whole blood methylomics and clinical data from 363 systemic lupus erythematosus (SLE) patients and 508 healthy controls (CTRLs). Transcriptomics data were analysed through a differential gene expression (DGE) analysis using the DESeq2 package. The serum and the urine metabolomic peaks change from mass spectrometry analysis were quantified and statistically tested. The peaks’ annotation was performed automatically comparing the m/z and retention time stored in the Ceu Mass Mediator database. The annotated metabolites were filtered based on the metabolites detected or predicted. The methylomic analysis and annotation were performed using the ChAMP R package. Common sources of variations among the -omics were identified by the Multi-Omics Factor Analysis (MOFA) integration. 3. Results Biomix carried out the analysis highlighting the most important features for each -omics, including its log2FC, p.adj and summarizing volcano plots. Furthermore, transcriptomics data analysis produced output ready for EnrichR and GSEA tools, to facilitate the exploration of altered biological processes. An option of the tool allowed to insert a panel of genes to define a subgroup of patients to compare to CTRL. This tool was used with the Precisesads database to focus on the subgroup of SLE IFN-α positive patients selecting a panel of 26 IFN-α genes. The reliability of the tool came from the results that highlighted the inflammatory environment and the hydroxychloroquine as one of the factors characterizing the SLE patients compared with CTRLs, as expected2,3. Moreover, a new suggestion on the nucleotide salvage pathway and lysophosphatidic acid signalling was highlighted. These features were corroborated by the unsupervised integration approach of MOFA, where they contribute the most to the factors that best separate SLE patients and CTRLs. 4. Conclusions: This user-friendly tool, based on R, can be launched for Linux and Microsoft OS. It proposes to make the multi-omics analysis accessible also to users not an expert in bioinformatics. The tool will allow to include soon single nucleotide polymorphism (SNPs) data analysis and integration, and other integration analysis tools such as iClusterPlus4 and Similarity Network Fusion5 to widen the user choices.

Nr:	17
Title:	Label-Free Single Cell Classification via Machine Learning and Optical Cell Signature Analysis
Authors:	David Dannhauser, Paolo A. Netti and Filippo Causa
Abstract:	Cells contain specific information, that can be useful for their classification as well as for a wide range of therapeutic purposes. Nowadays, to reveal potential cell information in a straightforward manner is still challenging. Human peripheral blood mononuclear cells (PBMCs) can be easily isolated from blood of healthy donors or buffy coats (a by-product from hospital Blood Banks) and therefore of valuable interest for diagnostic single cell approaches. Although PBMCs have a different composition, phenotype, and activation status than cells found in intestinal tissue, they can be seen as liquid biopsy of our body. For instance, liquid biopsy has drastically revolutionized the field of clinical oncology, offering the possibility of continuous monitoring by repeated sampling. In fact, clinicians need the appearance of different cell types and their status, for precise diagnosis. Nowadays, cell discrimination is routinely performed by cytometric analysis based on surface receptor expression, which is cost and time intensive. Alternatively, a label-free classification based on a liquid biopsy is highly demanded. Therefore, we present a simple microfluidic based single cell discrimination approach for by the mean of a label-free light scattering, together with a machine learning procedure for cell type prediction in liquid biopsies. Therefore, optical signatures (label-free information of morphological cell information) are recorded by a static light scattering apparatus (λ=632.8nm), which continuously measure scattering profiles of passing living cells. Obtained cell signatures are matched with pre-calculated simulation curves to retrieve the searched for morphological cell properties: dimension of cell and nucleus, optical density of nucleus and cytoplasm. Machine learning uses the morphological cell information as input to predict the searched for cell types. A prediction model was trained with more than 100 cells for a wide range of human peripheral blood: Erythrocytes, T- and B-lymphocytes, Natural Killer cells, Monocytes, Circulating Tumour Cells as well as M0-, M1- and M2-Macrophages. Separate cell types were diluted in viscoelastic cell medium and analysed. The separate measurement of different cell types allowed us to create a labelled training set of cell classes, which is needed for the machine learning based classification approach. However, we tested different classification algorithms, regarding single cell type classification accuracy. Moreover, we tested mixed cell samples to show the performance of our label-free classification approach. For instance, a classification accuracy above 90% was obtained using a fine Gaussian SVM-trained classifier. Furthermore, we show for the first time the ability to label-free predict unknown cells from scattering data using an open-set neural network approach, which significantly expand the application field of the presented single cell classification method. However, compared with standard flow cytometric approaches—which are known to have high instrumentation and service costs—our method is very simple and cost-effective, permitting a classification of cell subtypes without large numbers of cells and resource-intensive labelling. More importantly, measurements are realized using a lab-on-a-chip approach permitting the measurement of living cells in suspension, which furthermore are collectable and re-usable for other diagnostic investigations or therapeutic approaches.

Nr:	18
Title:	SurviveAI: Long Term Survival Prediction of Cancer Patients Based on Somatic RNA-Seq Expression
Authors:	Omri Nayshool, Nitzan Kol, Elisheva Javaski, Ninette Amariglio and Gidi Rechavi
Abstract:	Motivation: Prediction of cancer outcome is a major challenge in oncology and is essential for treatment planning. Repositories such as The Cancer Genome Atlas (TCGA) contain vast amounts of data for many types of cancers. Our goal was to create reliable prediction models using TCGA data and validate them using an external dataset. Results: For 16 TCGA cancer type cohorts we have optimized a Random Forest prediction model using parameter grid search followed by a backward feature elimination loop for dimensions reduction. For each feature that was removed, the model was retrained and the area under the curve of the receiver operating characteristic (AUC-ROC) was calculated using test data. Five prediction models gave AUC-ROC bigger than 80%. We used Clinical Proteomic Tumor Analysis Consortium v3 (CPTAC3) data for validation. The most enriched pathways for the top models were those involved in basic functions related to tumorigenesis and organ development. Enrichment for 2 prediction models of the TCGA-KIRP cohort was explored, one with 42 genes (AUC-ROC = 0.86) the other is composed of 300 genes (AUC-ROC = 0.85). The most enriched networks for both models share only 5 network nodes: DMBT1, IL11, HOXB6, TRIB3, PIM1. These genes play a significant role in renal cancer and might be used for prognosis prediction and as candidate therapeutic targets. Availability And Implementation: The prediction models were created and tested using Python SciKit-Learn package. They are freely accessible via a friendly web interface we called surviveAI at https://tinyurl.com/surviveai.

Nr:	20
Title:	Label-Free Sperm Classification Using Machine Learning Based Tracking in Microfluidics
Authors:	Luigi Fausto Canonico, David Dannhauser, Maria Isabella Maremonti, Claudia De Clemente, Paolo A. Netti and Filippo Causa
Abstract:	One out of five couples face infertility problems partially caused by environmental triggers and human lifestyle changes. A key factor for revealing spermatozoa functional characteristics is the investigation of spermatozoa motility and morphology with computer-assisted systems to give clinicians a fast and accurate screening of infertility parameters. In natural conception millions of sperm undergo a progressive physiological selection during their journey within the female reproductive system (FRS). During their competitive race, sperm interact with several biological checkpoints that allow the passage of competent sperm. In assisted reproduction techniques (ART), the success of embryo development depends mostly on the quality of gametes, and pregnancy outcome is strictly related to both the oocyte and the fertilizing sperm quality. Therefore, during ART, it is mandatory to select a mature and viable sperm with high DNA integrity. Although, sperm selection has been one of the primary focal points during the development of in vitro fertilization techniques, the possibility to non-invasively identify sperm endowed with these features is still an unsolved problem. Nowadays, ART is based on swim-up and density gradient centrifugation approaches and allows the recovery of fractions enriched with sperm with normal morphology and/or forward progressive motility. Although, such methods are those routinely used for in vitro sperm selection, it is becoming increasingly clear that new and more rigorous procedures should be considered. Recently, microfluidic technologies have emerged as a powerful tool that can closely replicate the in-vivo physiological conditions, overall in ART, while being able to achieve successful outcomes. We developed a versatile microfluidic device, which mimic the main guiding mechanisms and the morpho-physic barriers of the FRS and investigate motion patterns of free-swimming spermatozoa, offering the possibility to passively select functional and highly motile sperm cells exploiting a low-cost and non-invasive approach. Out of swimming patterns our self-written image processing routine automatically extract standard motion parameters, plus novel morpho-physic spermatozoa parameters. Due to the use of a supervised machine learning model combined with microfluidics, we can classify motion parameters of a raw semen sample in less than 30 min with a throughput rate above 1 000 cells h-1. The implementation of novel motion parameters for the supervised machine learning classification allowed us to classify immotile, in-situ, slow-progressive and rapid-progressive with an accuracy of more than 92%. In addition, our microfluidic concept allowed us to charge raw semen and analyse motion patterns at different stages of the platform. Such versatile microfluidic platform can significantly improve the study of male infertility by precise classification of single cells from real samples. Our preliminary results show the high potential of computer-assisted spermatozoa analysis, when machine learning, advanced microfluidics and image processing are combined. Our microfluidic measurement platform will be from significant interest for clinicians, which are interested in the precise classification of spermatozoa motion classes and the influence on assisted reproduction techniques.

Nr:	21
Title:	Study of Deformation-Dependent Cell Motions in Microfluidic Flows
Authors:	Maria Isabella Maremonti, David Dannhauser, Paolo A. Netti and Filippo Causa
Abstract:	In recent years, it has been demonstrated how cell deformability represents an important indicator of cell type, cycle, and state, becoming also a specific marker for a label-free microfluidic based separation. For example, it is well known that the in-flow dynamics of red blood cells are governed by cellular mechanical properties, such as internal viscosity and cytoskeleton elasticity. Thus, the nonlinear interplay between cell mechanics and flow may generate complex dynamics, which remain partially unexplored experimentally. In general, simulative and experimental results demonstrate that, non-spherical particles suspended in a fluid medium and subjected to an external flow field (e.g. shear flow), undergo complex 3D-motions by varying their orientations. For instance, when the revolution axis of the spheroidal object is in the shear plane, the object can rotate as a tumbling motion. Differently, when the long axis of the object oscillates around a mean orientation in the shear plane and the membrane rotates about the spheroidal shape, the object takes a tank-treading motion. Another in-flow dynamics is found when the object revolution axis is perpendicular to the shear plane, taking a rolling motion. In microfluidic systems, involving viscoelastic suspending fluids, the presence of fluid elasticity further complicates such dynamics, as recently demonstrated. In this work, we present a simple microfluidic chip for the label-free study of cell dynamics. The chip is designed to firstly align and secondly deform cells via tuneable viscoelastic forces before an asymmetric expansion of the channel cross-section where different cell dynamics are detectable, depending on individual cell parameters. In fact, we show how the in-flow deformation of cells influence their dynamic motion in-flow. We measured a set of in-flow parameters (orientation angle, aspect ratio, cell deformation and cell diameter) as a backward analysis of cell mechanical response. In particular, our experimental observations of breast cell lines show different CD and orientation angles changing upon the multiple cell states, as healthy or pathological. Thus, we specifically show how different ranges of CD values cause different cell motions describing distinct cell classes. For instance CD values above 0.15 enable tank-treading motions, while lower CD values facilitate tumbling. We conclude that, governing factors distinct motion regimes are i) the imposed velocity field, ii) the chosen fluid rheology, iii) the initial position of the object and iv) its shape and we demonstrate how different cell types respond to such imposed conditions with different dynamics, identified as rolling, tumbling and tank-treading. As result, our microfluidic chip with tuneable rheological fluid properties enables a simple and direct analysis of bright-field images useful for the label-free mechanical phenotyping of various cell lines.

Nr:	27
Title:	Experimental Correlation between Ultrasound Motion Tracking of the Radial Artery on the Wrist and Blood Pressure Measurements
Authors:	Federica Confalonieri, Alessandro Colombo, Marco Travagliati and Leonardo Baldasarre
Abstract:	Recent miniaturization and silicon integration of pMUT ultrasound devices [1] can enable blood pressure (BP) monitoring on wearable devices. It is therefore critical to develop an algorithm correlating the ultrasound (US) measurement to BP. In this work, we present the validation results of the measured Heart Rate (HR) and the correlation between the Artery Wall Motion (AWM) and the diameter of the radial artery with the difference between Systolic and Diastolic Pressure, DP. Ultrasound signals are acquired on the radial artery near the wrist using a commercial probe multiple times per day over a span of two days. AWM is then estimated with doppler axial autocorrelated motion tracking processing [2] and the reference BP and HR are measured using an ISO 81060-2:2013 BP monitor. The estimated HR with US matches well with the reference. Results collected on the same half day show good correlation with the DP but on the entire combined data the correlation is very low; this can be explained by the variability of artery stiffness over more than half day period. This shows that AWM alone is not sufficient to estimate BP and that also the artery stiffness should be evaluated to enable BP monitoring with US. REFERENCES [1] Jiang, X., Tang, H. Y., Lu, Y., Ng, E. J., Tsai, J. M., Boser, B. E., & Horsley, D. A. (2017). Ultrasonic fingerprint sensor with transmit beamforming based on a PMUT array bonded to CMOS circuitry. IEEE transactions on ultrasonics, ferroelectrics, and frequency control, 64(9), 1401-1408. [2] Jiang, X., Perrot, V., Varray, F., Bart, S., & Hartwell, P. G. (2021). Piezoelectric Micromachined Ultrasonic Transducer for Arterial Wall Dynamics Monitoring. IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, 69(1), 291-298.

Nr:	32
Title:	Radioimmunomodulating Property of Super-Paramagnetic Iron Oxide Nanoparticle (SPION) in vitro Used as Alternative Contrast Agent for MRI-Guided Liver Stereotactic Body Radiotherapy (SBRT)
Authors:	Alexander Kirichenko, Michael R. Shurin, Galina V. Shurin and Aishvarya J. Godla
Abstract:	Background. Ferumoxytol® (Feraheme, AMAG Pharmaceuticals, Waltham, MA) is a SPION agent that is increasingly utilized off-label as hepatic MRI contrast. This agent has the advantage of providing a functional assessment of the liver based upon its uptake by hepatic Kupffer cells proportionate to vascular perfusion, resulting in strong T1, T2 and T2* relaxation effects and enhanced contrast of malignant tumors, which lack Kupffer cells. The latter characteristic has been recently utilized for online radiotherapy treatment plan adaptation and more accurate MR-guided liver SBRT. However potential radiotoxicity of SPION has never been addressed for its safe use as an MRI-contrast agent during liver SBRT on MR-Linac. The purpose of this study was to determine radiosensitivity of human macrophages loaded with SPIONs in vitro. Methods. Human monocyte and macrophages cell line in cultures were loaded with clinically relevant concentration of Ferumoxytol (30µg/ml) for 2 and 24 h and irradiated to 3Gy, 5Gy and 10Gy. Cells were washed and cultured for additional 24 and 48 h prior to assessing their phenotypic activation by flow cytometry and function, including viability (Annexin V/PI assay), proliferation (MTT assay) and cytokine expression (Luminex assay). Results. Our results reveled that SPION affected both human monocytes and macrophages in vitro. Specifically, iron oxide nanoparticles decreased radiation-induced apoptosis and prevented radiation-induced inhibition of human monocyte proliferative activity. Furthermore, Ferumoxytol protected monocytes from radiation-induced modulation of phenotype. For instance, while irradiation decreased polarization of monocytes to CD11b+CD14+ and CD11bnegCD14neg phenotype, Ferumoxytol prevented these effects. In macrophages, Ferumoxytol counteracted the ability of radiation to up-regulate cell polarization to CD11b+CD14+ phenotype, and prevented radiation-induced down-regulation of expression of HLA-DR and CD86 molecules. Finally, Ferumoxytol uptake by human monocytes down-regulated expression of pro-inflammatory chemokines MIP-1α (Macrophage inflammatory protein 1α), MIP-1β (CCL4) and RANTES (CCL5). In macrophages, Ferumoxytol reversed the expression of IL-1RA, IL-8, IP-10 (CXCL10) and TNF-α, and up-regulates expression of MCP-1 (CCL2) and MIP-1α in irradiated macrophages. Discussion. This study defines the radiomodulating properties of SPIONs on human monocyte and macrophage cell lines exposed to 60Go gamma-rays within stereotactic body radiotherapy dose range. This has been validated using both cellular viability and proliferation efficiency assays. This study also finds strong evidence that SPIONs reversed the effect of radiation on the expression of pro-inflammatory cytokines involved in initiation and development of radiation-induced liver damage. Conclusion. SPION agent Ferumoxytol increases resistance of human monocytes to radiation-induced cell death in vitro and supports anti-inflammatory phenotype of human macrophages under radiation. The effect is radiation dose-dependent and depends on the duration of Feraheme uptake.

Nr:	33
Title:	The Association of Blood Cadmium with Glycated Hemoglobin in Population-Based Cross-Sectional Study: Insights from KNHANES Big Data
Authors:	Kisok Kim and Hyejin Park
Abstract:	Although there are a number of studies reporting that cadmium (Cd) exposure may be a potential risk factor for diabetes mellitus (DM), few nationwide epidemiological researches have analyzed the association between blood Cd concentration and glycated hemoglobin (HbA1c) profile as a marker of long-term glycemic fluctuation. The current population-based national study was conducted using Big Data built as a result of the Korea National Health and Nutrition Examination Survey (KNHANES). In the total sample of Korean adults, the HbA1c levels were 5.47% among participants in the lowest quartile of blood Cd level and 5.69% among those in highest quartile. The trend for the prevalence of a risky (> 5.7%) HbA1c level (prediabetes or diabetes conditions) was significantly related to an increased quartile blood Cd concentration. After adjustment for confounders, participants with the highest quartiles of blood Cd had increased odds of a risky (> 5.7%) HbA1c level (adjusted odds ratio = 1.44; 95% confidence interval, 1.17–1.78) compared with those with the lowest quartile of blood Cd. These results suggest the possibility that Big Data collected from a large population can be widely used in the health field, including evaluation of health effects of exposure to harmful substances.

Nr:	36
Title:	Effect of Ibuprofen on Red Blood Cells Studied Using Label-Free Digital Holotomography.
Authors:	Talia Bergaglio and Peter N. Nirmalraj
Abstract:	Understanding the interaction between over-the-counter (OTC) drugs and red blood cells (RBCs) is crucial for the assessment of dose-dependent drug-induced hematological adverse effects. However, it is challenging to monitor in a label-free manner, the real-time nanoscopic changes of RBCs as they interact with drug molecules. Here, we present a method for the label-free detection and quantification of ibuprofen-induced RBC shape changes with high spatial (~ 200 nm) and temporal resolution (~ 2s), based on digital holo-tomographic microscopy (DHTM). We demonstrate that refractive index (RI) tomograms of RBCs in the wet state can be used to extract and quantify the morphological (diameter, surface area, volume, thickness, sphericity) and chemical properties (hemoglobin concentration and content) of RBCs and to automatically identify different RBC morphologies using a machine learning (ML)-based classifier. We report the dynamic morphological transformation from normocytic RBCs to echinocytes upon introduction of ibuprofen molecules. Spicule formation and movement across the RBC membrane was recorded over a period of 20 minutes using DHTM and spicule morphology was further characterized at the nanoscopic level with atomic force microscopy (AFM). Echinocytosis was observed to be reversible at lower ibuprofen concentrations (0.25 mM and 0.5 mM) but the normocytic RBC morphology did not recover with higher ibuprofen concentrations (1.5 mM and 3 mM), as measured over a period of 1.5 hours. Additionally, results from molecular dynamics simulations further elucidate the interaction of ibuprofen molecules on RBC membrane at experimentally inaccessible timescales (nanoseconds). Our findings demonstrate the applicability of high-throughput microscopy and ML-based automated image analysis methods in the fields of digital pathology and personalized medicine as a hematology platform for the diagnosis of blood disorders and for monitoring the dose-dependent effects of prescribed and OTC medications in a cost-effective manner.

Nr:	37
Title:	Patient Condition Adaptive Path System (PCAPS) as a Clinical Decision Support Tool for Transplantation
Authors:	Hwa Hwa Chung, Sean Kong, Sean Kong and Wei Quan Teo
Abstract:	Post-surgical complications are common (7-15% undergoing major operations). These complications from post-transplant surgery pose a more significant problem due to the use of immune modulators. In most cases, post-surgical complications require immediate intervention. However, not all patients with the same disease are in the same state. Therefore, this added layer of complexity requires tailored management according to the patient's needs. There are many indications for liver transplants of varying levels of urgency. The end-stage liver disease (MELD) scoring model currently determines the urgency of a liver transplant but may fail to predict postoperative survival outcomes. A recent observational study by Amara et al. classified the different types of postoperative transplant complications into broad categories and timelines in which they occurred and highlighted areas for continual improvement. Building on this knowledge, it is evident that current clinical care pathways may not adequately fulfill healthcare tasks for different patient states and even less so for atypical patient states. These further emphasize the need for a patient-tailored post-surgical management system. Each postoperative transplant case is an important learning point, especially for those with adverse patient outcomes. Using the Clavien-Dindo classification system, we will sort each postoperative complication and its relevant clinical management. Furthermore, we propose using the rule-based algorithm to help to integrate clinical data patterns into each patient's condition. And with prescriptive analysis to predict the optimum management route. This system, the Patient Condition Adaptive Path System (PCAPS), comprises two major tools: Clinical Process Chart (CPC) and Unit Sheet (U.S.). The CPC allows a comprehensive overview of a patient's treatment. The Unit Sheet component incorporates details of medical therapy and is linked to the prescription system and the patient's medical record. Our observational study aims to generate a proof-of-concept (POC) model of the PCAPS. This model would convert structured and unstructured patient-level data from electronic Health Records (eHR) into algorithms for patient condition adaptive pathway generation. We aim to incorporate PCAPS into the standard of care management of transplantation. This will assist clinicians in making critical clinical decisions based on data points from the current patient state. It is important to note that this algorithmic system serves as a guide and does not look at replacing the discernment of clinicians.

Nr:	41
Title:	Physical Activity Variability is Associated with Increased Risk of Incident Atrial Fibrillation
Authors:	Minseon Park, Seung Min Lee, Ju Sam Hwang, Su Hwan H. Kim and Hyung-Jin Yoon
Abstract:	Obesity as well as weight variability is an established risk factor for many chronic diseases including atrial fibrillation. Most studies have used body mass index (BMI) measurements recorded for investigating risks of weight change to find this association. Few studies have taken into account physical activity (PA) changes that occur throughout the follow-up period, and there is no study investigating the effect of discontinuity in exercises, like PA variability, on AF development. Thus, the objective of this study was to investigate whether PA variability is a potential risk factor for incident AF. To examine the relationship between PA variability and AF, the data from 3,172,724 subjects (using the Korean National Health Examination Service), who had undergone health screenings ≥3 times and had no history of AF between 2009-2013, were retrospectively analyzed to find a risk of AF over 7 years (2014-2020). AF is defined by the code(I48) in ICD-10, and the combined total PA (MET-min/week) was calculated based on the International Physical Activity Questionnaire (IPAQ). PA variability for each individual was calculated using a coefficient of variation (CV). The risk of incident AF was analyzed with a Cox model (adjusted variables: age, systolic blood pressure, diastolic blood pressure, fasting glucose, total cholesterol, estimated glomerular filtration rate, average physical activity, BMI, smoking, drinking, and past medical history). A positive association was found in both men (adjusted Hazard Ratio 1.070; 95% confident interval, 1.048 – 1.092) and women (aHR 1.043; 95% CI, 1.015 – 1.071) with a 1 unit increase of CV of PA. High PA variability was significantly associated with AF development except for the low activity (<600 MET-min/week) group. In conclusion, PA fluctuation was associated with incident AF, especially in individuals who had performed moderate (600-3000 MET-min/week) or high PA (>3000 MET-min/week) during 2009-2013.

Nr:	42
Title:	Use of CMOS Image Sensor for Early Detection of Ischemic and Haemorrhagic Stroke
Authors:	Fabrizio Palma, Giuseppe Pignataro and Simone Paciotti
Abstract:	We present the development of a lab-on-chip system potentially able to determine specific miRNA levels that enable a differential diagnosis between ischemic and hemorrhagic stroke, through the specialization of CMOS Image Sensors. In particular, the system allows investigations on the photoluminescence of samples of biological liquid to be analyzed (plasma, lysate, biological fluid) following the capture of the specific miRNA by an antisense set of ad hoc designed Peptide Nucleic Acids (PNA) that confers the biological specificity and sensitivity. The CMOS Image Sensor-biochip is modified with a first PNA that captures the target miRNA. A second PNA bringing a fluorescent tag binds the target miRNA enabling detection of the 3-component complex by the CMOS.

Nr:	43
Title:	Manipulation of Virtual Objects Using Five-Finger Pressure of Handheld Haptic Controller in Virtual Reality Environment
Authors:	SangHun Nam
Abstract:	In this study, a method for manipulating a virtual object was studied by utilizing the pressure data of the five fingers of the handheld haptic controller when manipulating virtual objects in a virtual environment. The haptic controller under study can receive different pressure values from five fingers and is designed to control the force feedback per finger by using a linear motor matched to the five fingers. The virtual reality (VR) haptic controller is designed for right- and left-hand use and communicates data to the computer through Bluetooth. When a user grab the haptic controller with user’s fingers, the data measured by the pressure sensor are transmitted to the MCU, and it controls five linear motors to move them to the calculated position. The driving part uses five linear motors and is designed to withstand the force (22N) that adults press with their fingers, so that the position of the user's fingers can be maintained. When a user grabs a virtual object in virtual reality, the linear motor maintains the position when a virtual finger touches it, providing a sense of holding the virtual object. The five-finger buttons on the haptic controller can move independently to render virtual objects of various shapes and sizes. A control module was developed to control the haptic controller and create VR services and contents using the Unity game engine. In order to support a user interface using five fingers in VR, a VIVE tracker was attached and used to track the position and posture of the controllers. When the haptic controller is grabbed with fingers, the linear actuator moves in a linear motion, but a person grasps an object while moving the finger joints, and the hand in virtual reality must be visualized so that the finger joints move. It is necessary to transfer natural senses by mapping the characteristics of hardware sensors and human physical characteristics. The linear movement of the haptic controller and the finger movement of the virtual hand model are synchronized by mapping the position of the motor transferred from the haptic controller to the angle of the skeleton-based finger joint. In order to analyze the pressure value when the fingers hold the object while a person grips the object, a glove equipped with five pressure sensors was fabricated and the pressure data when the object was manipulated with five fingers was analyzed. Since the force applied to the finger varies depending on the object, experimental models with various shapes and weights were produced with a 3D printer, and pressure sensor data applied to the finger were obtained using the manufactured gloves. To implement interaction in VR, 36 spherical physical colliders were installed in the hand. The virtual hand was divided into six zones: thumb, index finger, middle finger, ring finger, small finger, and palm. Even within each area, the contact parts of the fingers are different according to the shape of the virtual object, so five physical colliders were installed on the thumb, six colliders on the other fingers, and seven colliders on the palm. Assuming that the user holds a virtual object with his hand when contact occurs in the three areas, the object is set to move together with the movement of the controller. In this study, we proposed a haptic controller capable of independently controlling five fingers and software related to virtual object manipulation, which can be used for hand and finger rehabilitation contents.

Nr:	45
Title:	The Lomb-Scargle Periodogram-Based Algorithm to Detect Differentially Expressed Genes Along Pseudotime
Authors:	Hitoshi Iuchi and Michiaki Hamada
Abstract:	Trajectory inference with single-cell RNA-sequencing (scRNA-seq) provides high-resolution representations of the biological process and contributes to understanding cell dynamics. Although many algorithms have been developed to infer trajectory and calculate pseudotime, only a limited number of algorithms have been released to detect differential gene expression patterns between two groups (e.g., wild-type vs. knock-out strain). In addition, most algorithms for identifying differential gene expression patterns do not support bifurcating trajectories. Therefore, we propose a Fourier transform-based approach to detect differentially expressed genes along the pseudotime between two groups. Although the general fast Fourier transform requires uniform sampling points, it is impossible to control uniform time points in trajectory analysis. Therefore, we developed an algorithm based on the Lomb-Scargle periodogram, which is used for periodicity analysis of data with non-uniform time intervals. We first evaluated the algorithm's performance on simulated data using the receiver operating characteristic curve. As a result, we showed that the accuracy of our approach is comparable to or better than existing algorithms. Moreover, numerical experiments also showed that our algorithm achieves high-level accuracy for data with non-continuous pseudotime due to the difficulty of trajectory inference. On the other hand, our approach is computationally more expensive than other methods because it employs a permutation test to calculate the p-value; it is within the practical range considering the time required to acquire scRNA-seq data. Furthermore, we applied our method to publicly available trypanosome scRNA-seq data and examined each method's trend. In summary, we showed that the Lomb-Scargle-based approach performs as well as or better than existing approaches using simulated. Additionally, we determine the preference for each method with trypanosome scRNA-seq data. The advantage of our method is compatibility for bifurcating trajectories and non-continuous trajectories. The Lomb-Scargle-based approach will contribute to understanding dynamic biological processes.

Nr:	47
Title:	Evaluation of Electromagnetic Susceptibility of Biopotential Electrodes for Long Term Monitoring
Authors:	Tiago Nunes, Ferran Silva, Mireya Fernández Chimeno, Marcos Quílez and Hugo Silva
Abstract:	As days go by, the number of devices that we have available to monitor our health keep increasing, allowing us to keep track of our blood pressure, heart rate or even performing ECGs in our homes. Of course, this means that the conditions in which the measurements are done are no longer controlled environments in controlled situations where the patient is in a hospital laying down while the measurements are being done. Now, he can be standing still or performing an activity, outside or at home. The number of scenarios in which the acquisition will be made is virtually unlimited. For that reason, the conditions in which the biosignal will be acquired may not be the most appropriate to extract a clear signal from where all the relevant information can be deduced. Besides all the motion artifacts that may affect the biosignal, the acquisition system itself, being an electronic equipment, is subject to electromagnetic interference which can in turn have a negative impact, direct or indirect, on the signal being acquired. This interference can couple to the circuitry of the system in many ways. It can couple directly to the circuit board of the system or to the electrodes connected to it. In this work we focus on the latter and we try to understand how the materials that compose the electrodes being used in an acquisition system will behave in the presence of electromagnetic disturbances. For this evaluation, a series of measures was done using a TEM cell, the EuroTEM. In this chamber, it is possible to generate a constant electrical field and consequently evaluate the electromagnetic susceptibility of a given device. In this case, we studied the behaviour of three different electrodes by placing them inside the TEM cell with a disturbing signal of a varying frequency starting at 10MHz up to 1GHz, with 1MHz steps. The three new electrodes evaluated are made of three different materials: silver, graphene, and carbon. All the electrodes are morphologically identical, extremely thin, flexible, and dry. They are intended to replace the usual Ag/AgCl gelled electrodes in long term monitoring of biosignals. The interference received by the electrodes were examined by connecting the latter to a spectrum analyser which registers the power received by the electrode for each frequency. This was repeated with every electrode for vertical and horizontal polarization. The results from these measurements show a response that varies with frequency and most of all with the material. In fact, there is clear evidence that silver electrodes are much more susceptible to an external electrical field than the electrodes made of carbon and graphene. Such a phenomenon can be explained by the higher conductivity of silver when compared to carbon and graphene. Furthermore, the high changes verified when the polarization of the field changes shows at what point the geometry of the electrode has an impact on susceptibility of the electrode to an external disturbing element.

Nr:	48
Title:	Automated Classification of Movement Quality in the Single Leg Squat Using Convolutional Neural Network Model
Authors:	Kyue-Nam Park and Sihyun Kim
Abstract:	Background: The single-leg squat (SLS) is a clinical movement test as a reliable assessment tool for the evaluation of the load of the anterior cruciate ligament, patellofemoral injury risk and lower extremity pain. Previous study used three inertial measurement units to classify the performance of SLS using random forest classifiers, achieving a 77% classification accuracy. However, the accuracy of SLS classification is proportional to the number of wearable sensors, and it is not suitable for clinical application of telehealth. Posture analysis using convolutional neural network (CNN) is suitable in telehealth platform because CNN model can track key point-locations in the image/video frame without wearable sensors and markers that can be used to calculate the angle of the joint during SLS. Purpose: Current study used pre-trained CNN to differentiate the movement quality in the SLS test using decision tree analysis, and to compare the frontal trunk, pelvis, knee, and summated angles among grades classified based on the criteria of the SLS quality test. In addition, current study used CNN algorithm to determine which factors were the most influential factors among the trunk, pelvis, knee, and summated angles. Methods: Participants performed SLS three times and recorded by a smartphone camera. 91 participants were classified as three groups (good, reduced, and poor grades) through movement quality of SLS test by experienced physical therapists. CNN algorithm was used to assess the frontal trunk, pelvis, knee, and summated angles when performing SLS. A one-way analysis of variance was used to compare the angle differences among three groups. Classification and regression tree (CART) was used to classify three groups and to identify the most influential factors among the trunk, pelvis, knee, and summated angles to the grades of SLS test. Results: The frontal knee angle and summated angle in the poor group were significantly larger than the reduced and good group. The frontal trunk angle in the poor group was significantly larger than the good group. CART analysis showed that the classification of the three groups showed 76.9% accuracy in the test model of the decision tree. Knee angles and summated angles were the most influencing factors to SLS movement quality. The cutoffs were a knee angle of 11.34° and a summated angle of 28.4° for the classification criteria. Conclusion: The movement quality in SLS test can be identified differences in the frontal trunk, knee, and summated angles, and can be classified using the knee and summated angles using CNN model without markers or equipment. Accordingly, smart devices with automated pose estimation model can be used to assess the joint angles and discriminate the movement quality of SLS test for remote clinical and sports application in telehealth platform in individuals with knee pain.

Nr:	50
Title:	Bioindicator to Monitor Environmental Humidity Inside Packaging
Authors:	Carolina M. Natal, Margarida Dias and Ana A. Roque
Abstract:	Most drugs are moisture sensitive. Humidity compromises drugs efficacy before reaching their shelf life, which can cause medication error due to altered dose, misleading results or change the appearance of the drug. There is a need to monitor humidity in packaging, especially in medicines and nutraceutics. Nowadays, desiccants are placed inside pill bottles to protect drugs from humidity. However, desiccants lose their properties after reaching a certain level of water adsorbed. Thus, it is relevant to monitor the desiccant humidity level inside the packaging to determine when the desiccant is saturated. In this work, we designed a bio-based multilayer material that changes its optical properties as a function of humidity. This bioindicator can produce a visible optical signal in the presence of humidity saturated environments. The signal obtained is easy to view by the user, and does not require an electronic equipment to produce the signal. The bioindicator components are widely available, inexpensive and pharma grade. The bioindicator presents low toxicity, is totally biodegradable and sustainable. The results showed that the bioindicator changes colour when the desiccant reaches about 20% of the initial weight, which means that when the desiccant is 20% saturated the bioindicator starts give an alert. The response of the bioindicator is dependent on the humidity percentage that it is exposed to. In an environment with 80% of relative humidity it takes 3 days to change the colour of the bioindicator, while in an environment of 60-70% of relative humidity it takes between 7 to 11 days. The work was developed within the scope of the “SMASUS - Smart and Sustainable Packaging” project (POCI-01-0247-FEDER-047007; LISBOA-01-0247-FEDER-047007), which was co-financed by Portugal 2020, under the Operational Program for Competitiveness and Internationalization (COMPETE 2020) through the European Regional Development Fund (ERDF).

Nr:	52
Title:	In-Vitro Study of the Application of Augmented Reality Headset HoloLens 2 for Dental Virtual Model Alignment on the Physical Dental Arch
Authors:	Mykolas Akulauskas, Vygandas Rutkūnas, Tomas Blažauskas, Eglė Eidukaitytė, Karolis Butkus and Darius Jegelevičius
Abstract:	Background: Augmented reality (AR) devices extend real world surrounding with addition of virtual information. Such a functionality is useful for dental surgeons then planning implantation for instantaneously control of the results. Placing a virtual dental model in real time on a real dental arch (model or real person) in the direct field of vision is a relevant and desirable task in dentistry. AR headset HoloLens 2 is a modern system with potential applicability for many areas. However, the application in medicine requires thorough studies of accuracy and usability. The objective of this study was to implement and to test in-vitro the application of AR technology for dental virtual model alignment on the physical dental arch. Materials and methods: AR headset HoloLens 2 with software, created using Vuforia (10.5) library and Mixed Reality Toolkit in Unity (2020.3.17) programing environment was used. Digital maxillary dental model was used for the experiments. Physical model was additively manufactured using resin-based 3D printer. For calibration procedure, pencil-like positioning probe with attached 30x30x30 mm cube for tracking was modelled and printed. For tracking physical maxillary dental model tracker in a form of “thick L” of size 30x40x20 mm (dimensions of enclosing cuboid) was printed and attached to the dental arch using dental wire and silicon. Both positioning probe and tracker cubic surfaces were covered with a non-repeating contrasting pattern. Two experiments: calibration and hologram to physical model superimposition accuracy analysis, estimating trueness and precision, were conducted. In calibration experiment digital dental model was superimposed on its identical counterpart of 3D printed dental model. Superimposition was performed with the probe touching physical dental surface parts with intention to link digital dental model with physical dental model’s marker. Calibration accuracy was assessed with the use of recorded characteristic point’s coordinates. Hologram accuracy analysis was performed after calibration in static (AR headset on the head model) and dynamic (AR headset on the user) settings. In both settings videos of dental hologram and tracked characteristic points were recorded. Extracted characteristic point’s coordinates were used to assess holograms accuracy. Results: Calibration results showed overall accuracy and precision surpassing 1 mm threshold in distance related measurements. Hologram accuracy dependency on different AR virtual model alignment settings is noticeable in both types of measurements: using video records and using characteristic points. Due to calibration procedure visible hologram positional offsets (5.53-8.39 mm) from physical model is seen in acquired videos. The left position (opposite to attached dental model tracker) in dynamic setting showed lower precision of distance measurements reaching standard deviation 1.46-1.6 mm than other positions in both settings (ranging 0.01-1.04 mm). Conclusions: Newly created marker showed only satisfactory results regarding its accuracy in dental application. Additional studies with appropriate experimental environment are needed especially for improving dynamic virtual model alignment.

Nr:	53
Title:	Straight Channel Microfluidics for Viscoelastic Exosome Separation: Influence of Geometry and Materials
Authors:	David Poustka, Anna Paříková, David Kramoliš, Jaromír Havlica and Jan Malý
Abstract:	Exosomes, 30−200 nm membrane vesicles secreted by almost all mammalian cells, have been a hot topic in recent years due to their valuable cargo of information, such as proteins, microRNAs and DNAs, and their promising role as biomarkers. Equally discussed are methods of their separation that could replace ultracentrifugation with its many drawbacks. Inertial microfluidics utilising viscoelastic separation in non-Newtonian fluids with sheath flows are promising candidates. However, since factors such as channel geometry, fluid viscosity and flow rate all heavily affect final separation, proper designing and application of these straight microfluidic channels turn out to be not as straightforward. Therefore, we developed a precise numerical model of all main forces acting on exosomes in straight microfluidic channels, performed experiments with fabricated chips of various channel dimensions and identified an optimal approach to designing and utilising these channels for exosome separation in several key applications. Also, since PDMS may pose an issue due to its deformability at higher pressures leading to alteration of channel geometry, we simultaneously fabricated chips from a rigid UV and heat cured thermoset polymer Ostemer and compared resulting separation efficiencies. We believe that viscoelastic microfluidics may serve as an interesting alternative to conventional exosome separation techniques.

Nr:	55
Title:	Classification of Mammogram Patches Using Convolutional Neural Networks
Authors:	Adam Mracko and Ivan Cimrak
Abstract:	Nowadays, artificial intelligence methods are experiencing great expansion in medical fields. Convolutional neural network models are predominant in medicine because of their suitable properties for image classification. In our work, we considered mammographic breast images. The mammograms were obtained from the CBIS-DDSM (Curated Breast Imaging Subset of Digital Database for Screening Mammography) database. Our experiments mainly worked with the mammogram patches, where the finding was located.The experiments aimed to compare the accuracy and confusion matrices of various models with different hyperparameter values, and different processing of mammograms into patches. The patches were processed in different ways, with variable resolutions or different sizes of the surrounding area of the important finding. The bulk of the experiments focused on binary classifiers that worked only with calcifications or masses. We compared the effects of using pre-trained and not pre-trained architecture weights on the ImageNet dataset. We also observed the impact of oversampling and class weighting methods on resolving the unbalanced dataset.

Nr:	56
Title:	Relationships Among Accessibly of Health Information, Habits, and Mental Health in Elderly: A Sustainable Action Plan for Resident-Centered Regional Health Promotion
Authors:	Shuhei Oba, Takeo Shibata, Akemi Kunimatsu, Chika Hiraga and Yoko Shimizu
Abstract:	Introduction Yuzawa town, located in the Niigata prefecture of Japan, is famous for its hot springs and ski resorts. A sustainable action plan for resident-centered regional health promotion (Yuzawa family health plan) was initiated in 2002. Surveys have been conducted every five years. An accessibly of health information has been increased recent years in elderly. This study aims to evaluate relationships among accessibly of health information on the internet, habits, and mental health status in elderly. Methods The survey of Yuzawa family health plan was conducted in 2017. Questionnaires were sent residence’s homes randomly. The questionnaires were composed with living alone, living with marriage partner, living with son or daughter, living with grandchildren, living with brother or sister, health consciousness, attention to health, health concern, awakening in a morning, clockwork life, fitness habits, smoking, taking alcohol, anxiety, loneliness, consultation, resolve suffering, economic problem, handle stress, taking a rest, risk breaching, communication, life satisfaction, talking with family, job at home, intercommunion with friends, greeting with neighbors, relationship with neighbors, relationship with babies, relationship with elementary school children, relationship with junior high school students, relationship with high school students, relationship with young adults, relationship with middle aged adults, and relationship with elderlies, occupations, good sleep, taking breakfast, nutritional balance, mastication status, confirmation of food compositions, exercise habits, yearly health check-up, dental check-up, brushing teeth habits, The survey was also conducted anonymously. Every participant consented the study. Chi square tests were used to evaluate relationships among accessibly of health information, habits, and mental health status in elderly (over 65 years old). This study was approved by the ethics committee of Tokyo Women's Medical University. Results 435 elderly residents joined the survey in 2017. Men showed higher accessibly of health information than women (p<0.001). Accessibly of health information related with living alone (p=0.001), living with son or daughter (p<0.001), talking with family (p=0.001), relationships with young adults (p=0.002), consultations (P=0.007), anxiety (p=0.011), taking alcohol (p=0.016), health concerns (p=0.020), solving problems (p=0.022), demining difficulties (p=0.026), relationships with neighborhood (p=0.048). Discussions Elderly residences with health concerns showed lower accessibly of health information (11.6%). Residences who could not solve their problems and demine difficulties also showed lower accessibly. Most residences with health concerns could not have health information on the internet. Yuzawa town is also famous with Japanese liquor. Higher drinking rates are shown in elderly (65.8% for men and 31.5% for women). Though drinkers showed higher accessibly of health information than non-drinkers, 82.0% of drinkers did not have health information. Some drinkers had interests in their health and got health information on internet. Numbers of elderly living alone have been increased from 2002. Health promotion programs giving health information for elderly with health concerns and difficulties is needed.

Nr:	61
Title:	A Bayesian Network-Based Framework to Uncover the Causal Effects of Genes on Complex Traits Based on GWAS Data
Authors:	Yaning Feng, Hon-Cheong So, Liangying Yin and Yaning Feng
Abstract:	Deciphering the relationships between genes and complex traits could help us better understand the biological mechanisms underlying diseases. Univariate gene-based analyses are widely used to characterize gene-phenotype relationships. However, they are subject to the influence of confounders. On the other hand, while some genes directly contribute to traits variations, others may exert their effects through other genes. How to quantify individual genes’ direct and indirect effects on complex traits remains an important yet challenging question. We presented a novel framework to decipher the total and direct causal effects of individual genes using imputed gene expression data from GWAS and raw gene expression from GTEx. The study was partially motivated by the quest to differentiate “core” genes (genes with direct causal effect on the phenotype) from “peripheral” ones. Our proposed framework is based on a Bayesian network (BN) approach, which produces a directed graph showing the relationship between genes and the phenotype. Since the approach could uncover the overall causal structure, we can directly examine the role of individual genes and quantify the direct and indirect effects by each gene. An important advantage and novelty of the proposed framework is that it allows gene expression and disease trait(s) to be evaluated in different samples, significantly improving the flexibility and applicability of the approach. Also, it is extendable to decipher the overall causal network of more than 2 traits. We verified the feasibility and validity of the proposed framework by extensive simulations and applications to 52 traits in the UK Biobank (UKBB). The identified (direct) causal genes were found to be significantly enriched for genes highlighted in the OpenTargets database, and the enrichment was stronger when compared to conventional univariate gene-based tests. Besides, many enriched pathways were supported by the literature. Some of the enriched drugs have been tested or used to treat patients in clinical practice. Our proposed framework provides a novel way to prioritize genes with high causal or direct causal effects and estimate the ‘importance’ of such genes.

Area 1 - Scale-IT-up

Nr:	5
Title:	How to Lower Socioeconomic Inequalities in Health with Digital Therapeutics? Healthification: A Short Animation
Authors:	Tobias Kowatsch
Abstract:	Non-communicable diseases (NCDs), including common mental disorders, impose enormous health burdens on individuals and lead to substantial economic challenges (Jacobson et al., 2022; Kowatsch & Fleisch, 2021). Known risk factors relate primarily to a lifestyle characterized by tobacco and excessive alcohol consumption, physical inactivity, or an unbalanced diet. This lifestyle can lead to obesity, hypertension, and cardiovascular and neurodegenerative diseases. Unfortunately, individuals with lower socioeconomic status (SES) are substantially more affected by NCDs (Mackenbach et al., 2008; Wang & Geng, 2019). These individuals are also underrepresented in clinical and non-clinical trials (Davis et al., 2019; Ford et al., 2008). As a result, health interventions are potentially only effective for individuals with higher SES and do not address those most in need. Therefore, it is essential to understand how to reach and engage individuals with lower SES. To this end, we propose to “hijack” the comfort zones of vulnerable individuals (e.g., television shows and social media) and frame digital therapeutics so that they are primarily scalable (e.g., by using everyday technology, such as smartphones), unobtrusive (e.g., by using digital biomarkers), relatable (e.g., with conversational agents and family members as intervention components), and enjoyable (e.g., by gamified approaches), with health effects being the side effects. A short animation has been produced for the scientific community to promote this perspective and to rethink digital therapeutics aiming to lower socioeconomic inequalities in health (Anonymous 2022).