Neural-Symbolic System for Predicting COVID-19 Positivity

Thanks to the huge amount of data collected by hospitals, it is now possible to exploit Machine Learning (ML) to build predictive models that can learn from data for identifying medical pathologies. The potential of Deep Learning (DL) and ML algorithms are well known but, in a field such as medicine, it is necessary to build interpretable and explainable systems instead of black-box systems as the de facto in DL. This work applies those techniques to both clinical data and Computed Tomography (CT) scans to predict COVID-19 positivity. To achieve an explainable model, we used both neural systems, for classifying and analyzing CT scans images, a symbolic model, Decision Tree, for analyzing clinical data concerning patients and a Neural-Symbolic architecture that integrates both systems using Hierarchical Probabilistic Logic Programming (HPLP). Experiments confirm that the proposed system provides a prediction accuracy of almost 90% and is able to provide explanation of the classifications.