一种基于无监督的医学图像器官分割模型预训练方法
By using unsupervised pre-training methods to augment data and optimize the loss function with unlabeled data, the generalization problem of medical image organ segmentation models is solved, and the training effect and speed of the models are improved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HEFEI RAYCISION MEDICAL TECHNOLOGY CO LTD
- Filing Date
- 2023-12-26
- Publication Date
- 2026-07-17
AI Technical Summary
Existing weakly supervised and semi-supervised learning methods have limited generalization ability in organ segmentation of medical images due to the size of the dataset, which affects the effectiveness and speed of the segmentation results.
An unsupervised approach is used to pre-train a medical image organ segmentation model. Data augmentation is performed using unlabeled training data, and the model is optimized using contrast loss, classification loss, and reconstruction loss functions, including a combined structure of encoder, predictor, and decoder.
This improved the training efficiency and speed of the medical image organ segmentation model, avoided the limitation of labeled dataset size, made full use of unlabeled data for pre-training, and obtained optimized weight parameters.
Smart Images
Figure CN118015396B_ABST