Brain mri image nuclei segmentation model training and segmentation method, device and medium
By generating initial pseudo-labels through brain map registration and combining them with closed-loop iterative training of a deep learning model, the problems of dependence on manual labeling and insufficient registration accuracy in existing technologies are solved. This achieves high-precision, low-cost brain nucleus segmentation and improves the model's generalization ability and stability.
Patent Information
- Application Number
- CN202610742933.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-05-27
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies rely heavily on manual annotation in brain nucleus segmentation, resulting in limited registration accuracy and a lack of closed-loop optimization mechanisms, making it difficult to achieve high-precision and stable segmentation.
By image registration of brain templates with nucleus structure annotations to brain MRI images, initial pseudo-annotations are generated, a training sample set is constructed, and a deep learning segmentation model is used for training. A closed-loop iterative mechanism of prediction-update-retraining is established to gradually improve the quality of pseudo-annotations and the segmentation accuracy of the model.
It significantly reduces the reliance on manually labeled data, achieves high-precision and stable brain nucleus segmentation, improves the model's generalization ability and robustness, and can effectively utilize unlabeled data, reducing data preparation costs and time consumption.
Smart Images
Figure CN122415585A_ABST
Abstract
Citation Information
Patent Citations
Cell instance segmentation method and apparatus based on pseudo-label dynamic updating
CN114708286B
Online learning enhanced atlas-based auto-segmentation
US10410348B2