The Challenge:
- Manufacturer of leading FISH imaging platform required robust nuclei segmentation and abnormality risk estimation modules for their integrated software
- Tightly clumped and overlapping cells
- Precise interpretation is necessary to overcome ambiguity, avoid false positives, and detect all rare events
Our Approach:
- Customized our state-of-the-art deep learning based segmentation technology for client dataset image characteristics
- Created robust segmentation & marker assignment algorithms for overlapping cells
- Implemented custom abnormality risk measure
Results:
- Nuclei detection: 98% precision, recall, and F1 score
- Nuclei segmentation: 0.98 mean Dice score
- Key marker detection: 95% precision, recall, and F1 score
