Binghao Chai

This study tests if stain normalization improves retrieval with pathology foundation models, finding no consistent benefit and potential harm from Macenko normalization.

Pheno-MYCN links slide-level MYCN prediction to interpretable morphological sub-populations in neuroblastoma, revealing a continuous gradient of MYCN-associated features.

Across 14 sarcoma and mimic subtypes, six pathology foundation models beat ResNet50 but lost accuracy under multi-institutional staining; adding 121 externally stained slides narrowed the gap.

The paper introduces Multi-SpinX, an AI framework that extends SpinX to track mitotic spindles and kinetochores in multicellular environments across different mitotic stages.