Siamese bayesian networks for AI based differential diagnosis

Abstract

Differential diagnosis refers to the process of differentiating between two or more conditions which share similar signs or symptoms. Classical methods such as Bayesian Networks proposed in the past to automatically obtain a differential diagnosis do not consider negative evidence for prediction and also lack the ability to model hidden influences on diseases. In order to address the shortcomings of the existing methods for automated differential diagnosis, we propose a novel Siamese Bayesian Networks that takes into consideration the absence of a symptom as a strong negative evidence to converge to the actual diagnosis. We show that the proposed algorithm has a 40% improvement over manual differential diagnosis of disorders and a 10% improvement over classical Bayesian Networks approach for differential diagnosis.