BIOSTEC_DC 2022 Abstracts

Full Papers

Paper Nr:	1
Title:	Support System for the Diagnosis of Vocal Pathologies
Authors:	Joana Fernandes
Abstract:	The aim of this work is to develop a system to help diagnose speech pathologies associated with the larynx. For this, an Alpha system will be developed, which will be put into operation in clinics, so that specialists collect signals from patients with diagnosed pathologies, since it is necessary to increase the signal database, with more subjects and more vocal pathologies. The best set of acoustic parameters will be investigated in order to improve the classification, as well as the use/optimization of Deep Learning tools, based on convolutional networks, Transfer Learning techniques and recurrent networks. Finally, the Beta system will be developed, which will be developed through improvements to the Alpha system, through contact with professionals, the set of parameters that contribute with more discrimination capacity and those with Deep Learning tools that obtain better accuracy.

Paper Nr:	2
Title:	Analysis of Speech Signals in Patients with Schizophrenia
Authors:	Felipe Teixeira
Abstract:	Schizophrenia is a chronic and severe mental disorder with heterogeneous presentations. An early diagnosis of a mental illness helps in providing accurate treatment and in the recovery of patients. The changes in schizophrenic speech are currently conceptualized as a component of negative symptoms between negative symptoms has five domains, where is include blunted affect, asociality, alogia, anhedonia and avolition. In this domains, speech acoustic changes are reflected in two domains- blunted affect (diminished expression of emotion) and alogia (poverty of speech). My proposal PhD thesis entitled ”Analysis of speech features in subject with schizophrenia”, initially, an analysis of parameters will be carried out in order to apply them to machine learning tools, such as neural networks, and thus obtain a classification of the subject under analysis. With the conclusion of this work, the software is expected to aid medical decision-making through the evaluation of the severity of the pathology.

Paper Nr:	3
Title:	Data Platform for the Unification and Analysis of Extracellular Vesicle Data
Authors:	Hannah Janout
Abstract:	Extracellular vesicles (EVs) are membranous structures naturally shed or produced by cells. They act as cell to cell communicators and significantly influence cell functionality, hence affecting a body’s overall behavior. Thus, they offer a direct way of determining the (patho)physiological state of cells and can be used as a diagnostic tool. In past studies, EVs have been shown to indicate certain kidney diseases and cancer types. However, through their potential in the medical field, processes for the cultivation, modification, and evaluation of EVs are highly sought after with no international standard available. Consequently, the data accumulated during these processes vary greatly, complicating the comprehensibility and exchange of data. In this paper, we present the concept of a data platform used to securely store EV data accumulated from various experiments and provide algorithms for their analysis. By introducing a NoSQL database, a uniform structure for EV experiments is created, simplifying their analysis and interchangeability. The included algorithms combine conventional data and image analysis methods with evolutionary strategies and neural networks, enabling a diverse analysis of incoming data and solving several EV research challenges. Furthermore, stored EV data can be compared to new EV samples extracted from patients. Thus, correlations between illnesses and EV behavior/properties and further diagnostic evaluations are made. Therefore, unlocking the EVs’ full potential as a diagnostic and therapeutic tool. As a basis for the planned data platform, two preliminary projects are also described. The first workflow focuses on detecting and tracking EVs and quantifying contained green fluorescent proteins (GFP) on a fluorescence microscopy image series. The second workflow focuses on the analysis and quality assessment of EVs through multimodal imaging.

Paper Nr:	4
Title:	Automated Cell Segmentation for Micropatterning Microscopy Images
Authors:	Jonas Schurr
Abstract:	In this paper, we present an automated framework including cell segmentation for quantitative analysis of fluorescence microscopy-based images from protein micropatterning experiments. The goal of such experiments is to identify and quantitate protein-protein interactions in the cell membrane and even in the cytosol of living cells. Therefore, we developed a fast and easy-to-use workflow for μ-patterning analyses. In this publication, we describe a method for the segmentation of patterned cells. It is based on an Unet for the automated segmentation of the patterned cells. Furthermore, an evolutionary strategy allows the estimation of the pattern on the cells. Based on the pattern on the extracted cells and their intensities, additional information about the characteristics of the protein-protein interactions is extracted automatically. As we show in this paper, by additional automated cell segmentation we further automatize the information extraction and provide comprehensive micropatterning analyses.

Paper Nr:	5
Title:	Heuristic Domain Shift Adaptation for the Analysis of Blood Vessel Images
Authors:	Andreas Haghofer
Abstract:	In assisted diagnostic, image segmentation methodologies represent one exciting way of supporting medical experts. Due to various imaging setups, the usability of these segmentation models is often limited due to the incompatibility with the current imaging setup. As an alternative to existing methods, we present an approach that does not require any change of the used segmentation model. We apply an automated preprocessing workflow to the used image data, extending the compatibility of pre-trained image segmentation neural networks for images, otherwise leading to unusable segmentation results. Our algorithm incorporates heuristic optimization of preprocessing steps and generates a specialized preprocessing workflow for each newly arriving image dataset without the need to adjust the segmentation neural network itself. In the current state, our workflow is already capable of processing images of blood vessels to be segmented by a neural network, which leads to an improved segmentation quality compared to the segmentation on the raw images.