Bloodstream infection related septic shock multi-modal fusion prediction method and system

Abstract

The application provides a blood stream infection related septic shock multi-modal fusion prediction method and system, belonging to the field of intelligent medical treatment, and solving the problems of long time consumption, insufficient sensitivity and insufficient multi-modal information analysis capability in blood stream infection related septic shock. The method comprises: acquiring multi-modal clinical data of a target patient suspected of blood stream infection within a preset observation time window to obtain a multi-modal multi-scale feature sequence formed by combining each modality spatiotemporal feature vector sequence; inputting the multi-modal multi-scale feature sequence into a time sequence encoding network based on a multi-head attention mechanism to obtain an aligned multi-modal time sequence representation sequence; setting a learnable modality weight parameter for each modality, and updating the learnable modality weight parameter, the encoder and the time sequence encoding network through a task loss function according to the label of whether blood stream infection related septic shock occurs or not in the training process to obtain a multi-modal fusion model for prediction.

Images