Advancing medical prognosis based on graph concepts and artificial neural networks

Abstract

Deep learning (DL) has proven to succeed in a wide range of domains, ranging from acoustics and images to natural language processing. However, applying deep learning to graph-like data is a non-trivial task because of the unique properties of graphs. Recently, a significant amount of research efforts have been devoted to this area, notably improving graph analyzing techniques. Employing these novel techniques in the medical domain is still unprecedented, but with great potential for contributions when analyzing Electronic Medical Records (EMRs) using both graphs and artificial neural networks. In this context, this project aims to provide for the computational phenotyping by relating the entities found in EMRs, narrowing causes, symptoms, procedures, and treatments of one or many diseases more accurately than previous methods and also with more reliability, conciseness, and self-interpretability though semi-supervised hierarchical clustering on Graph Neural Networks (GNNs). The activities related to such a task include dealing with the various issues related to the application of deep learning to the specific context of clinical data using graphs. These issues include pre-processing demands to produce large and cleaned datasets of clinical data; modeling complex information through graphs; fine-tuning DL architectures concerning the specific problems of prognostic care; iterative training-testing rounds to achieve highly accurate methods; and, clinical validation of the results. The intend results, however, are of broad interest to the national and international scientific community. (AU)