DeepGEM 2.0: 1-Minute AI Pre-screening for All Approved Lung Cancer Targeted Drug Genes

KingMed Diagnostics and the First Affiliated Hospital of Guangzhou Medical University jointly released DeepGEM 2.0, a multimodal AI model that pre-screens all gene targets corresponding to domestically approved lung cancer targeted therapies from a single digital pathology slide in as little as one minute. "Use AI at the top; make it affordable, simple, effective and safe at the grassroots," said Zhong Nanshan, Chinese Academy of Engineering academician, at the 10 October 2026 AI Empowering Precision Lung Cancer Diagnosis Summit. Many county- and prefecture-level hospitals lack next-generation sequencing, forcing patients to choose between chemotherapy and traveling to tertiary centers for genotyping. DeepGEM 2.0 enables local stratification: high-probability predictions get confirmed via widely available PCR or immunohistochemistry nearby, while low-probability cases skip full NGS scanning. The 2025 v1.0 (Tencent-partnered) was upgraded to 2.0 on 10 October 2026 after nearly a year of multi-center validation on approximately 8,000 digital pathology slides from 31 Chinese provinces. DeepGEM 2.0 ingests routine H&E slides and outputs both gene-mutation probability and spatial tissue distribution. New genes can be added within 12-16 weeks of a new drug's approval, keeping the panel in sync with cutting-edge therapies. The model is lightweight-deployed on standard PCs via KingMed's KMDP digital pathology system. Future deployment extends to over 500 medical institutions. A multi-center prospective randomized controlled trial and health economics study led by Professor Liang Wenhua's team will evaluate real-world performance. The combined "AI pre-screening plus targeted verification" model is expected to bring lung cancer gene testing costs down to several hundred yuan, replacing much more expensive NGS panels. KingMed's national telepathology network and a three-tier collaborative network mark DeepGEM 2.0's transition from performance validation to large-scale clinical application validation.