Multimodal Fusion Models For Pulmonary Embolism Mortality Prediction

This article presents the multimodality fusion strategies and models for the task of PE risk stratification. Clinical variables taken from the patients’ EHRs, such as demographics, vital signs, chief complaint, clinical history, inpatient and outpatient medications, past medical history, and laboratory test results, are used in combination with CTPA scans to create a fusion model that offers better prognosis designations than imaging data alone. This fusion model is more optimal for integration into the clinical workflow due to its end-to-end approach and high sensitivity, which helps reduce false-positives and clinical alert fatigue.