Session 1 :
Technologie, Robotique et
Date: Friday 2nd October 2020. 10:30-12:00
10:30-10:50: Leveraging Blockchain Technology to Enhance Pharmaceutical Supply Chain Management. Shayma Romdhani, Issam Nouaouri, Jihene Tounsi, Hamid Allaoui and Said Gattoufi.
Abstract: Amyotrophic Lateral Sclerosis (ALS) is a fatal neurodegenerative disease with limited treatment options. Finding reliable prognosis models remains a key issue in ALS research. This work presents a simple patient prognosis model for survival where result uncertainty is taken into account. Patient data are reduced and projected onto a 2D space using Uniform Manifold Approximation and Projection (UMAP), a novel non-linear dimension reduction technique. A total of 4 906 patients are included in the study, aggregating past clinical trials as development data, and real-world population data as validation data. Predictors used are age, sex, onset location, time since onset, baseline weight, functional loss, and estimated functional loss rate. UMAP projection of patients shows an informative 2D data distribution. As data availability hinders complex model design, the projection is divided into three zones with relevant survival rates. These rates are defined using confidence bounds: high, intermediate, and low 1-year survival rates at respectively 92% (+/- 4%), 83% (+/- 4%) and 62% (+/- 4%). Future patient 1-year survival is estimated using zone membership. This approach requires a limited set of features, is easily updated, improves with additional patient data, and accounts for results uncertainty.
11:00-11:20: Measuring the expression of RNA sequences in hundreds of RNA-seq libraries at nucleotide resolution. Mariam Bouzid, Daniel Gautheret and Rayan Chikhi
Abstract: This research aims at combining the latest machine learning and big data analytics techniques with the healthcare systems domain knowledge to explore the added value of the medical data.
11:30-11:50: Clinical Laboratory Optimization. Sohrab Faramarzi Oghani, El Ghazali Talbi and Eric Varlet.
Abstract: As medical databases grow larger and larger, medical experts often lack appropriate and accessible tools to make the best of the datasets available and transform data into actionable information. Many knowledge extraction algorithms provide relevant results but fail to provide explainable and transparent results. Accountability is paramount in healthcare, and hospital staff cannot rely on black box tools when it comes to taking informed decisions. To address this situation we propose an algorithm able to structure thousands of electronic medical records by similarity and typicality. Using a rank-based approach suitable for high-dimensional data, we associate each patient's record to a very similar yet more typical record. This provides a structure suitable for data visualization, allowing for both a high-level summary of a cohort and its representative patients, and a detailed representation of similarities and relations in each cluster. We applied this method to electronic medical records of diabetic patients, providing an easy tool for visualization and exploration of these data with the added benefit of explainability.