AI-Based DeepGEM Tool for Predicting Gene Mutations in NSCLC Patients: A Randomized Controlled Study

Application of the Artificial Intelligence-Based Gene Mutation Prediction Tool DeepGEM in Patients With Non-Small Cell Lung Cancer (NSCLC): A Prospective, Multicenter, Randomized Controlled Trial

This prospective, multicenter, randomized controlled trial aims to evaluate the clinical utility of DeepGEM, an artificial intelligence (AI)-based mutation prediction tool based on histopathological whole-slide images, in patients with non-small cell lung cancer (NSCLC). The study will assess whether DeepGEM can facilitate molecular testing, increase targeted therapy utilization, and improve survival outcomes in a real-world clinical setting. Patients with stage II-IV treatment-naïve NSCLC and qualified pathology slides for DeepGEM analysis will be enrolled. Eligible participants with AI-predicted EGFR, ALK, or ROS1 mutations will be randomized in a 4:1 ratio to either the DeepGEM-informed group (clinicians can access AI results to guide further testing and treatment) or the standard care group (clinicians are blinded to AI results and follow routine care).

Inclusion Criteria:

• Age between 18 and 75 years, inclusive, at the time of enrollment.
• Histologically or cytologically confirmed non-small cell lung cancer (NSCLC) with clinical stage II-IV as per the 8th edition of the AJCC staging system.
• Availability of qualified histopathological whole-slide images that can be reviewed through the KindMED system(DeepGEM).
• Successful mutation prediction of EGFR, ALK, or ROS1 by the DeepGEM AI tool.
• No prior systemic anti-cancer therapy, including chemotherapy, targeted therapy, or immunotherapy.
• Willing and able to comply with study requirements, including follow-up and treatment; written informed consent must be provided.

Exclusion Criteria:

• Prior systemic anti-tumor therapy (chemotherapy, radiotherapy, targeted therapy-including but not limited to monoclonal antibodies or tyrosine kinase inhibitors) before enrollment.
• Failure of DeepGEM analysis or unqualified histopathological image quality.
• History of any other malignancy within the past 5 years, except for adequately treated basal cell carcinoma of the skin or in situ carcinoma (e.g., cervical carcinoma in situ).
• Cognitive or psychological barriers to understanding or accepting AI-based prediction or molecular testing.
• Pregnant or breastfeeding women, or women of childbearing potential who are not using effective contraception.
• Any other clinical condition that, in the opinion of the investigators, may interfere with the study protocol or compromise participant safety, including poor compliance with study procedures.