Neuro-Symbolic AI for Supporting Chronic Disease Diagnosis and Monitoring

In remote areas or regions with limited access to medical specialists, there is often a high reliance on telemedicine and Artificial Intelligence (AI)-based diagnostic tools. However, misdiagnoses or inadequate care may occur if the AI system lacks domain knowledge, failing to adhere to medical protocols. Despite the incredible research efforts applying AI in medicine, only a few models have been routinely adopted in medicine, due to issues related to trustworthiness. To address these concerns, Symbolic Knowledge Injection (SKI) has been proposed as a solution: it integrates domain-specific expertise into Machine Learning (ML) models, to improve their predictive capabilities. Despite their promising results in other fields, applicability of SKI in healthcare scenarios has not been thoroughly investigated, yet. Accordingly, in this study, we explore the applicability of a SKI method on medical datasets to evaluate: (i) how the predictive capabilities of ML models changes, (ii) their adherence to the medical protocols, and (iii) their robustness w.r.t. data degradation. Results demonstrate the potential of integrating data-driven models with established medical guidelines by improving different clinically relevant metrics.