AI-HEAT: A Clinical Decision Support System for Pediatrics Febrile Conditions

In recent years, several diagnostic challenges have developed due to the COVID-19 pandemic, including the post-infectious sequelae multisystem inflammatory syndrome in chil-dren (MIS-C). This syndrome shares several clinical features with other entities, such as Kawasaki disease (KD) and endemic typhus, among other febrile diseases. Endemic typhus, or murine typhus, is an acute infection treated much differently than MIS-C and KD. Early diagnosis and appropriate treatment are crucial to a favorable outcome for patients with these disorders. To address these challenges, a Clinical Decision Support System (CDSS) designed to support the decision-making of medical teams can be implemented to differentiate between these disorders. We developed and evaluated a CDSS based on a Triplet Loss Siamese Network to distinguish between patients presenting with clinically similar febrile illnesses, KD, MIS-C, or typhus. We used eight clinical and laboratory features typically available within six hours of presentation. The performance assessment for AI-HEAT, Logistic Regression, Support Vector Machine, XGBoost, and the TabPFN machine learning models was performed by computing Balanced Accuracy. AI-HEAT is a CDSS capable of obtaining performance similar to a state-of-the-art Transformer-type deep learning model such as TabPFN, with advantages such as being almost a thousand times smaller.