BIOSTEC_DC 2023 Abstracts

Full Papers
Full Papers
Paper Nr: 5
Title:

A Tailored Internet of Things Lighting Solution to Support Circadian Rhythms and Wellbeing of People Living with Dementia
Authors:

Kate Turley
Abstract: Dementia is a disease with no cure which affects around 50 million people worldwide. As such, it has become of shared interest to develop solutions which support the wellbeing and improve the quality of life for those impacted by the condition. An established method for achieving this is through the use of lighting; particularly circadian lighting. This type of lighting replicates sunlight for the indoors, as it changes in colour temperature and intensity throughout the day. Moreover, this dynamic lighting can regulate the body’s circadian rhythm, which is responsible for managing our mood, hormone balance, body temperature, sleep-wake cycles and rest-activity patterns. Therefore by better synchronising this rhythm for people living with dementia, it becomes possible to alleviate some of the symptoms associated with dementia, such as agitation, disrupted sleep cycles and low mood. This research focuses on a study designed to assess the impact this type of lighting has on both the circadian rhythm and the related wellbeing factors of people living with dementia. In order to achieve this, sensing devices are deployed which are able to track daily activity. This intelligent network of luminaires and sensors then functions as a feedback system, whereby the activity profile of an individual over time will provide data-driven insights which inform the actuation of the circadian lighting. The colour, intensity, timing and duration of the lighting will depend on the observed sensor data and therefore accommodates the needs of any person’s circadian rhythm. This tailored approach is state-of-the-art. In addition, the metrics generated by the sensing devices are available on a dashboard, meaning it is possible to view the health metrics in live time. This therefore provides an aid to the support network of people living with dementia. The final aim of this research is to understand the relationship between lighting, circadian rhythms and dementia, in order to better inform tailored lighting output in future. This then offers a better support to wellbeing than any ambient static lighting would.
Paper Nr: 6
Title:

Implementation and Optimization of Explainable and Trustworthy Artificial Intelligence Algorithms for the Analysis of Radiological Images
Authors:

Camilla Scapicchio
Abstract: Despite the obvious potential of Artificial Intelligence (AI) applications in medical imaging analysis, there are still many critical issues and limitations to overcome, mainly related to the availability of data, ethics, and the need for interpretability and generalizability, to effectively provide clinicians with reliable AI-based decision support tools. The research problem I’m aiming to address in my Ph.D. program is the investigation of possible solutions for these issues and limitations. In particular, I will address the main crucial limitation that currently prevents the real adoption of these systems in clinical practice, which is the lack of transparency and interpretability. In the first part of my Ph.D. program, I focused on COVID-19 applications, since COVID-19 represented a perfect use case to develop specific methods related to these topics, for the current social and clinical relevance of this pathology, and the consequent availability of multicentric public data. I worked on three different challenges, with the common thread of developing AI algorithms that are explainable and reliable. In every single application, I tried to cover a different aspect of interpretability, such as the possibility of explaining why the models predict what they predict, the possibility to have an analysis based on intelligible features, and the development of robust models. Good results in terms of the performance and explainability of the developed AI systems on the specific applications of COVID-19 have already been obtained. However, the aim is to extend these frameworks to other pathologies and imaging techniques.
Paper Nr: 7
Title:

Characterizing Intraoral 3D Scanners: Phenomenological Model Approach
Authors:

Mykolas Akulauskas
Abstract: Intraoral scanners (IOS) qualitative analysis is a prominent research topic. However, due to the wide range of clinical situations, it is hard to picture overall individual IOS accuracy since each of the cases produces different accuracy values. Although attempts by individual researchers and standards issuers have been made to approach this problem, there is still no general solution to solve this problem. In this research work, a model which could highlight IOS scanner’s accuracy in different clinical situations was proposed. The model consists of similarly shaped objects: one representing scanned and the other handmade object. Features representing IOS are extracted from scanned object and embedded with handmade, which results in a synthesized model capable to represent IOS in different situations. As a conceptual representation of this research problem, experiments with roundness of 90-degree angle edge as a feature for scanner representation was chosen.
Paper Nr: 8
Title:

Stress or No Stress? This Is the Question: Life-Like Stress Detection in Different Domains Using Different Hardware Solutions
Authors:

Josefine Welk
Abstract: The burden of stress at work or in everyday life plays a major role for many people. Based on this, the PhD project focuses on the investigation of different hardware solutions and domains with the goal to identify suitable devices and to develop a machine learning model for stress detection. In contrast to the partially already integrated stress detection of the devices, the model should not require a person-specific learning phase making it immediately applicable in everyday situations and requiring no user action to identify stress. For this purpose, a concept was developed that divides the project into a laboratory phase and a field phase. The laboratory phase focuses on the evaluation of different hardware solutions from the point of view of stress detection in a controlled environment. The result of the laboratory phase is a set of tested devices as well as a machine learning model that is used to label medical measurement data into "stress" and "no stress". The field phase, on the other hand, does not take place in a controlled environment, but evaluates the devices already identified as suitable from the laboratory phase in everyday situations. In addition to the devices, the model is also evaluated again and a comparative model is developed. The result of the field phase is a set of devices suitable for the detection of stress as well as a machine learning model for stress detection without a person-specific training phase. Currently, the project is in the phase of preparing the laboratory phase.
Paper Nr: 9
Title:

Simulation of Medical Shock Waves and Research on the Safety of Shock Wave Application on Contraindicated Tissue
Authors:

Hannah Janout
Abstract: Extracorporeal shock wave therapy (ESWT) has been used in regenerative medicine for many years. However, while this application's biological aspects are considered well-researched, their physical sides are still partially unknown. Therefore, parameters used for shock wave therapy are often taken from ultrasound therapy despite their fundamental differences. Additionally, it is conventional to use static descriptions for medical shock waves, although stochastic variations highly influence generation and propagation. In order to eliminate the deficiencies, real-life shock wave treatments are to be mimicked in computer simulations and analyzed with machine learning and white-box modelling techniques to investigate the stochastic influences of shock wave therapies. Thus, a probabilistic description of medical shock waves will succeed, with the help of which biological reactions and physical parameters can be correlated, to come one step closer to a standard for medical shock waves and improve regenerative ESWT. Furthermore, through computer simulation and consequent analysis, the application of ESWT on lung tissue is used to explore new applications of regenerative ESWT. The presence of air has always been considered a contraindication to ESWT on lung tissue, as high negative maximum pressure can cause pulmonary capillary haemorrhage. However, animal trials suggest that applying shock waves on lung tissue may be safe under certain conditions. In this paper, the author describes their PhDs thesis objectives in detail and provides insight into the first steps of simulating real-life shock wave treatments and their analysis to increase the application of regenerative ESWT.
Paper Nr: 10
Title:

Canine Lymphoma Categorization Using Deep Learning and Advanced Image Processing
Authors:

Andreas Haghofer
Abstract: The acquisition and analysis of histological sections form the basis for classifying lymphoma types necessary for providing suitable treatment for each animal. However, this is very time-consuming because the pathologists have to analyze the images manually. Besides others, one criterion to distinguish between different types of lymphoma is the nuclei size which is used for the presented workflow. By combining an artificial neural network with advanced processing algorithms, the proposed workflow can differentiate between three types of lymphoma in dogs by analyzing the histological image samples with a test accuracy of 88%.
Paper Nr: 11
Title:

3D Microtumours by Microfluidics: Fabrication and Applications
Authors:

Xingyun Yang
Abstract: Conventional two-dimensional (2D) cell culture is well-established but suffers from disadvantages, such as limited cell-cell and cell-extracellular matrix interactions, resulting in cell flattening or cell remodelling.1 Three-dimensional (3D) cell culture offers a more physiologically relevant tissue model.2, 3 We have fabricated 3D microtumours (3D-MMT) by a microfluidic technique in which tumour cells are encapsulated in viscous gels (e.g. Matrigel, collagen, agarose). ~100 3D-MMTs, of uniform-size and -composition, can be prepared in just 3 minutes. The diameter of the 3D-MMTs can be adjusted (300 to 900 µm), various shapes can be prepared (sphere, ellipsoid), and the cellular contents can be controlled. The 3D-MMTs can be cultured for up to 8 weeks. In an example application, 3D-MMTs containing ovarian cancer cells were treated with first-line chemotherapeutics (carboplatin, paclitaxel). The 3D-MMTs offered higher drug resistance than 2D cultures. Cells remaining after chemotherapy, minimal residual disease (MRD), can reinitiate the tumour growth.4 RNAseq analysis demonstrated that MRD 3D-MMTs recapitulated all 5 gene signatures found in samples from patients, while only one signature was present in 2D-cultured MRD cells. Further, a cytotoxicity assay showed that MRD 3D-MMTs are more sensitive to fatty acid oxidation inhibitors than treatment-naïve 3D-MMTs. In conclusion, batches of hundreds of 3D-MMTs can be generated, and can be used as a realistic model for rapid drug screening.