A naukri.com initiative
Arxiv
2d
50
Image Credit: Arxiv
HER2 Expression Prediction with Flexible Multi-Modal Inputs via Dynamic Bidirectional Reconstruction
- Current HER2 assessment models for breast cancer typically focus on analyzing H&E or IHC images separately despite the clinical need for their combined interpretation.
- Proposed adaptive bimodal framework enables flexible HER2 prediction using single- or dual-modality inputs via dynamic branch selection.
- Innovations include a branch selector for single- or dual-modality prediction, a bidirectional GAN for context-aware reconstruction, and a hybrid training protocol.
- Framework boosts H&E prediction accuracy from 71.44% to 94.25% and achieves 95.09% accuracy for dual-modality prediction, maintaining 90.28% reliability with only IHC inputs.
- Design allows for near-bimodal performance without synchronized acquisition, aiding resource-limited settings through reduced IHC costs.
- Experimental validation shows 22.81%/12.90% accuracy improvements over baseline models, with enhanced F1-scores for cross-modal reconstructions.
- System dynamically routes inputs for reconstruction or fusion pathways, mitigating performance issues from missing data while maintaining computational efficiency.
- The architecture demonstrates potential for widespread use in precise HER2 assessment across various healthcare environments.
Read Full Article
3 Likes
For uninterrupted reading, download the app