基于时空联合检测的心脏疾病分类模型构建方法
By preprocessing and reorganizing multi-lead ECG signals using a self-supervised learning method to generate pseudo-labels, and combining this with a pre-trained deep feature extraction network, the problems of insufficient multi-lead ECG datasets and data imbalance were solved, thus improving the performance of the heart disease classification model.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ZHEJIANG UNIV
- Filing Date
- 2022-12-30
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies suffer from a lack of multi-lead electrocardiogram datasets and data imbalance, and most algorithms rely on fully supervised methods, resulting in limited performance.
A self-supervised method based on generative learning is adopted. Pseudo-labels are generated by preprocessing and reorganizing multi-lead ECG signals. A deep feature extraction network model is built for joint pre-training. The model parameters are optimized by using pseudo-labels and pre-training loss functions. Then, fine-tuning is performed on the classification network model to reduce the dependence on labeled data.
It effectively reduces the reliance on labeled data and improves the performance of the heart disease classification model, especially in the case of insufficient data, thus enhancing the performance of the model in downstream tasks.
Smart Images
Figure CN115881308B_ABST