Systems and methods for metastatic cell detection and risk prediction
Graph learning methods using graph neural networks and contrastive learning effectively detect and predict metastatic cells, addressing the challenges of rarity and similarity, enhancing diagnostic accuracy and treatment strategies.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- OHIO STATE INNOVATION FOUND
- Filing Date
- 2025-11-20
- Publication Date
- 2026-05-28
AI Technical Summary
Conventional methods struggle to accurately detect metastatic cells due to their rarity and similarity to other cells, leading to delayed diagnosis and ineffective treatment strategies.
A computer-implemented method using graph learning techniques, including graph neural networks and knowledge-aware contrastive learning, constructs heterogeneous graphs from cell and gene profiles to identify metastatic indicators, such as precursor cells and occult metastatic cells, by integrating large-scale electronic health records and genetic data.
Enhances true positive identification of occult metastatic cells, reduces false positives, and predicts metastatic risk with improved precision by tracing metastatic transitions and identifying robust biomarkers, facilitating timely therapeutic interventions.
Smart Images
Figure US2025056344_28052026_PF_FP_ABST