Publications

A deep-learning system for gastric-cancer detection in histopathology whole-slide images, trained on pixel-level–annotated H&E slides and validated on 3,212 real-world WSIs (≈100% sensitivity, 80.6% mean specificity). Deployed at the Chinese PLA General Hospital and validated at two further hospitals. Calvin was a lead ML engineer on the deep-learning pipeline and a co-author.

A weakly-supervised framework for histopathology image segmentation that learns from image-level labels via instance-level pseudo-labeling, reducing the need for dense pixel annotation. Calvin led the state-of-the-art baseline experiments, contributed to the formulation, and presented the work at ICCV 2019 in Seoul.