Language driven segmentation base model for general medical image analysis
By using a language-driven intelligent segmentation model and pre-training with image-mask-description triples, the problem of visual cue dependence in existing technologies is solved, enabling efficient and accurate medical image segmentation for non-expert users, and applicable to various medical image types.
CN122312481APending Publication Date: 2026-06-30THE HONG KONG UNIV OF SCI & TECH
View PDF 0 Cites 0 Cited by
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- THE HONG KONG UNIV OF SCI & TECH
- Filing Date
- 2025-12-30
- Publication Date
- 2026-06-30
Smart Images
Figure CN122312481A_ABST
Abstract
This invention provides a full-stack medical imaging foundation model for performing medical image segmentation, classification, and localization using a unified language-image correlation mechanism. This invention leverages pre-trained language-guided associations to automatically identify and localize disease targets without requiring medical imaging knowledge or manual bounding box prompts. This invention enables comprehensive clinical tasks across multiple disease categories and various medical imaging techniques, improving segmentation efficiency and reducing the burden of manual input. By utilizing the operability of plain text prompts from non-radiologists, this invention improves real-world applicability and enables the deployment of foundation model-based segmentation in diverse clinical settings.
Need to check novelty before this filing date? Find Prior Art