Utilizing Explainable AI and Biosensors for Clinical Diagnosis of Infectious Vector-Borne Diseases

Abstract

Listen

Translate article

Infectious Diseases (ID) are a significant global threat due to their epidemic nature and substantial impact on mortality rates. COVID-19 has proven this assertion by wreaking havoc on human wellness and healthcare resources. This has underscored the need for early ID diagnosis to restrict the spread and protect human lives. Recently, Artificial Intelligence (AI)-assisted biosensors have shown great potential to assist physicians in making decisions to minimize mortality rates. However, their adoption in clinical practice is still in its infancy, primarily due to the challenges faced by physicians to interpret decisions derived from these black-box systems. The objective of this study is to earn the trust of physicians to promote their acceptance and widespread adoption in healthcare. Against this backdrop, this research is a pioneering effort to investigate not only the diagnostic accuracy of several Machine Learning (ML) algorithms for ID but more specifically how to leverage the benefits of Shapley values to provide valuable insights regarding the contribution of clinical features for early ID diagnosis. This analysis examines four ML algorithms that stem from different theories, such as Random Forest Classifier (RFC), Gradient Boosting Classifier (GBC), Support Vector Classifier (SVC), and Multilayer Perceptron (MLP). The visual analysis results presented for local and global interpretation facilitate the observation of the marginal impact of each clinical feature on a patient-by-patient basis. Therefore, the results of this study are expected to aid practitioners in better evaluating the diagnostic decisions of the ML models developed and boost the use of AI-assisted biosensors for ID diagnoses.