An interpretable automated detection system for FISH-based HER2 - PowerPoint PPT Presentation

MIDL 2020 – Short Paper An interpretable automated detection system for FISH-based HER2 oncogene amplification testing in histo-pathological routine images of breast and gastric cancer diagnostics Sarah Schmell, Falk Zakrzewski, Walter de Back, Martin Weigert, Uwe Schmidt, Torsten Wenke, Silke Zeugner, Robert Mantey, Christian Sperling, Ingo Roeder, Pia Hoenscheid, Daniela Aust and Gustavo Baretton

Complex Diagnostics and Increasing Workload FISH HER2 amplification testing AI assisted Pathologist Pathologist 2 2 2 3 2 3 2 4 1 3 5 + 4 3 3 2 3 4 3 2 2 # HER2 2 per nucleus 2 2 # CEP17 3 < 2.0 → HER2 negative → Not HER2-targeted therapy 2 ≥ 2.0 → HER2 positive low 2 → HER2-targeted therapy > 5.0 → HER2 positive high 1

1. Step: Nucleus Filtering with the Nucleus Detector Schmidt et al. (2018) Output (~80 × 80 px) Input (1600 × 1200 px) Segmentation of detected Nuclei HER2 CEP17 2

2. Step: Grading Nuclei with the Nucleus Classifier Postprocessing Input ( ~ 80 × 80 px) Output (Grading and CAM) (image-wide Grading) Artifact Background HER2 low amplification HER2 normal expression Artifact HER2 low amplification HER2 high amplification … 3

3. Step: Second Opinion from the Signal Detector Lin et al. (2017) Postprocessing Input ( ~ 80 × 80 px) Output (Grading and BBoxes) (image-wide Grading) Artifact HER2 low amplification HER2 normal expression HER2 low amplification Artifact HER2 low amplification HER2 high amplification … 4

4. Step: Reviewing the Report Enhancements to the AI assistance 5

THANK YOU An interpretable automated detection system for FISH-based HER2 oncogene amplification testing in histo-pathological routine images of breast and gastric cancer diagnostics