This prospective multicentre diagnostic study enrolled patients with clinical stage IA lung adenocarcinoma from three hospitals in China and used smartphone-captured surgical resection images to develop SuRImage, a deep learning model for intraoperative assessment. The model was tested for identifying invasive lung adenocarcinoma, diagnosing adenocarcinoma in situ/minimally invasive/invasive disease, and grading invasive lung adenocarcinoma. SuRImage showed AUCs of 0.84, 0.87, and 0.85 for these three tasks in the Guangdong Provincial People’s Hospital cohort, and the study reports better diagnostic performance than frozen section analysis. The authors concluded that SuRImage could help surgeons make more timely and precise intraoperative decisions in stage IA lung adenocarcinoma.