Cross-modal data analysis method based on ECG and CMR data

• Main Authors: HUANG ZHENGXING, WU FEI, DING ZHENGYAO
• Format: Patent
• Language: Chinese; English
• Published: 16.08.2024
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; DIAGNOSIS; HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATIONTECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING ORPROCESSING OF MEDICAL OR HEALTHCARE DATA; HUMAN NECESSITIES; HYGIENE; IDENTIFICATION; INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTEDFOR SPECIFIC APPLICATION FIELDS; MEDICAL OR VETERINARY SCIENCE; PHYSICS; SURGERY

The invention discloses a cross-modal data analysis method based on ECG and CMR data. The method comprises the steps that an ECG and CMR pairing data set is acquired; constructing a deep learning model; performing first-stage pre-training on the deep learning model, and processing ECG data and CMR data by adopting a self-supervision method and a supervised method respectively; second-stage contrast training is carried out on the deep learning model, contrast learning between ECG and CMR data is carried out, CMR encoder parameters are frozen, only an ECG encoder is trained, alignment between two data features is optimized, and fine adjustment is carried out on the ECG encoder through supervised learning; defining a total loss function; using hyper-parameter search, adopting different hyper-parameter combinations on the training set for training, verifying the performance on the verification set, selecting the optimal hyper-parameter combination, and testing the deep learning model adopting the optimal hyper-pa