Method and system for brain tumor segmentation based on missing modality MRI using a full modality training framework reuse
By reusing the full-modal training framework, the model is divided into a Common model and a Barebone model. The representation and segmentation are optimized by aligning with the homologous mask, which solves the robustness and stability problem of MRI brain tumor segmentation under missing modalities and achieves high-efficiency segmentation performance under limited complete modal samples.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-05-14
- Publication Date
- 2026-07-17
AI Technical Summary
Existing multimodal MRI brain tumor segmentation methods exhibit performance degradation under missing modal conditions, struggle to be trained collaboratively with a large number of missing modal samples under limited complete modal sample conditions, and suffer from inconsistencies between generated results and real data distribution.
A full-modal training framework reuse method is adopted, dividing the model into a Common model and a Barebone model. The Common model reuses the full modality learning capability during the training phase, while the Barebone model handles the missing modalities during the inference phase. Co-optimization of representation-level constraints and segmentation supervision is achieved through homogeneous mask alignment, thereby improving robustness and generalization ability.
This improved the clinical applicability and deployment stability of the missing modality MRI brain tumor segmentation model, reduced the distribution bias caused by pseudo-missing samples or pseudo-modal generation, and enhanced the model's stability and generalization ability under multiple combinations of real modal missingness.
Smart Images
Figure CN122415649A_ABST