基于人工智能的多模态脑卒中风险筛查模型的临床应用OA

Objective To develop an artificial intelligence-based multimodal model for stroke risk screening,and to improve the accuracy and efficiency of stroke screening.Methods With reference to the mainstream models like the"8+2"model of the National Stroke Prevention and Control Committee,the Prediction for ASCVD Risk in China(China-PAR)model,and the China Kadoorie Biobank(CKB)model,a multimodal model for stroke risk screening and preven-tion was established by integrating multi-dimensional data including blood biochemistry,clinical diagnosis,individual medical history,signs,and living habits.This model was applied in the clinical verification of 6 511 individuals who par-ticipated in stroke risk screening in Shaoxing People's Hospital from January to December 2023,and the above models were compared in terms of the consistency of prediction results and the acceptance rate among physicians.Results The consistency rate of risk classification was 71.31%between the self-developed model and the China-PAR model and 70.80%between the self-developed model and the CKB model,which were significantly higher than the consistency rates between the CKB model and the China-PAR model and between the"8+2"model and the other models.The consistency rate of risk classification was 89.79%between the self-developed simplified model(with 10 biomarkers)and the complete model(with 17 biomarkers),suggesting that the self-developed simplified model is suitable for primary outpatient screen-ing or large-scale physical examination.In addition,the self-developed model further improved the accuracy of risk assess-ment through percentage conversion and refinement of risk levels.Conclusion The intelligent screening model based on multimodal fusion can effectively improve the accuracy and timeliness of stroke screening,thereby providing a feasible so-lution for large-scale stroke screening.