Fig. 8: Performance comparison of multimodal survival prediction modules.
From: Vision-language model for report generation and outcome prediction in CT pulmonary angiogram
a Concordance index (C-index) for different combinations of modality inputs, including PESI scores, imaging (Img), clinical variables (Clin), diagnosis (Dia), and generated report text (Text), across the BUH and JHU testing cohorts. Multimodal fusion models consistently outperform unimodal baselines. b Decision Curve Analysis (DCA) of multimodal survival prediction modules illustrates the net clinical benefit of various unimodal and multimodal models across a range of threshold probabilities for risk stratification. c Kaplan-Meier survival curves for high-risk and low-risk groups stratified by the median risk score from the four-modal fusion model (Img + Clin + Dia + Text).
