AI Model Predicts Rare Gene Mutations in Patients With Lung Adenocarcinoma

By: JNCCN 360 Staff
Posted: Friday, October 10, 2025

A recent study published in JCO Clinical Cancer Informatics by Peiling Yu, MD, of the Department of Pathology, the First Affiliated Hospital of China Medical University, Shenyang, China and colleagues demonstrated the potential of an artificial intelligence (AI)–based model that uses hematoxylin and eosin (H&E)–stained histopathology slides to predict rare gene mutations in patients with lung adenocarcinoma. The model, which achieved levels of speed and accuracy that exceeded those of previous AI models used in tumor pathology, has the potential to expedite diagnosis and enable more personalized therapy for patients with non–small cell lung cancer (NSCLC).

“The artificial intelligence model developed in this study has a high accuracy in predicting gene mutations in lung adenocarcinoma, which is conducive to improving the management of patients with lung adenocarcinoma and promoting precision medicine,” the authors stated.

Study Design

The researchers developed the model using the ResNeXt101 framework, which employs an end-to-end learning method, eliminating the need for manual extraction of heuristic features.

The model was trained and validated using a dataset of 213 digitized H&E–stained slides compiled from two independent patient cohorts. Cohort 1 included 144 patients who underwent gene testing between 2018 and 2022 at the First Affiliated Hospital of China Medical University. Cohort 2 included 69 patients from The Cancer Genome Atlas–Lung Adenocarcinoma dataset.

The model was designed to predict mutations in six target genes—ALK, HER2, KRAS, RET, MET, and ROS1. Pathologists annotated, segmented, and denoised the tumor regions in the slides, after which each whole-slide image was subdivided into 224 × 224–pixel tiles to optimize computational efficiency and accuracy. After the preprocessing stage in which nontumor regions were removed, the cases were divided into training (70%), validation (10%), and test (20%) sets.

Results

Performance was evaluated using standard classification metrics, including area under the curve (AUC), accuracy, precision, recall, and F1 score. The ResNeXt101 model demonstrated high predictive accuracy in both internal and external validations. In cohort 1, an AUC range of 0.93 to 1.00 was reached, while external validation on cohort 2 demonstrated AUC values between 0.85 and 1.00 for five of the six genes. When applied to an independent dataset of metastatic lung adenocarcinoma, the model exhibited a good performance, with AUC values between 0.72 and 0.80 for HER2, KRAS, and RET mutations.  However, the study’s small sample size and the rare mutation set suggest that larger multi-institutional datasets are warranted to enhance the generalizability of the model and validate its real-world applicability.

Conclusion

Gene mutation profiling is critical for guiding targeted therapy in NSCLC, but current molecular diagnostics and next-generation sequencing can be costly and time-consuming. According to the authors, their AI approach has the potential to bridge the gap between histopathology and molecular testing, identifying likely mutation carriers directly from standard slides.

“Our model is capable of evaluating the status of key biomarkers through [whole-slide images], significantly shortening the time interval from diagnosis to the initiation of targeted therapy. It can assist pathologists by highlighting samples that may harbor potential mutations, allowing for prioritized examination,” they concluded.

Disclosure: For full disclosures of the study authors, visit ascopubs.org.

JCO Clinical Cancer Informatics

---