Ant Afu and the Fourth Hospital of Hebei Medical University develop DeepComp AI model to predict postoperative complications and survival risk in gastric cancer before surgery

A collaboration between Ant Group Afu Medical AI Lab and The Fourth Hospital of Hebei Medical University (Hebei Tumor Hospital) has produced three gastric cancer models: DeepComp for preoperative moderate-to-severe complications and long-term survival risk, RSA for early recurrence, and RCSA for liver metastasis. One paper was accepted by Annals of Oncology and another appeared in Nature Communications.

DeepComp draws on enhanced CT and clinical data, extracting tumor morphology, a 5 mm peritumoral microenvironment, and body composition at the L3 level. Across 11 centers and 5,237 patients, internal validation AUC was 0.888, with nine external cohorts ranging from 0.824 to 0.869. The team reports it outperformed nine existing clinical scores and ten surgeons of varying seniority in a human-machine comparison. They built separate models for each clinical scenario, not a general one.

Original sources (Chinese)