Artificial intelligence-based algorithm for predicting outcomes in early-stage lung cancer: An annotation-free imaging artificial intelligence study

Read full text Summarise with VIRAT
Authors: Keiju Aokage, Jumpei Ukita, Mamoru Miura, Keisuke Ogaki, Tomoko Kataoka, Noriko Mitome, Mitsuhiro Isaka, Masaya Yotsukura, Shun-ichi Watanabe, Masahiro Tsuboi
Cited by: 0
References: 32

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Objective: Surgery remains the standard treatment for clinical stage I non-small cell lung cancer (NSCLC). Conventional prognostic factors are often subjective and variable, highlighting the need for objective prediction. We developed and validated an annotation-free artificial intelligence (AI) model using computed tomography (CT) images and clinical data to predict prognosis in stage I NSCLC. Methods: In step 1, an AI algorithm was developed to predict pathological classifications from CT and clinical data collected in 3 prospective multi-institutional trials. In step 2, the model was refined and validated using a cohort from the National Cancer Center Hospital East. Models were trained to predict 5-year disease-free survival and overall survival. Performance was evaluated by sensitivity, specificity, predictive values, and receiver operating characteristic area under curve (AUC). Results: We analyzed 1217 patients in step 1 and 1338 in step 2. Models integrating CT imaging and clinical data outperformed models using either dataset alone. The pathology prediction model achieved an AUC of 0.787. The highest performance for 5-year disease-free survival (AUC = 0.757) and overall survival (AUC = 0.756) was obtained by combining preoperative clinical information, physician CT assessments, and an AI-based CT model. Adding AI outputs to clinical factors improved risk stratification, better separating high-risk from low- to intermediate-risk groups. Conclusions: Annotation-free AI models that integrate CT imaging with clinical data provide accurate, objective prediction of recurrence and survival in stage I NSCLC, complement conventional diagnostics, and support personalized multidisciplinary treatment planning.

Fields of Study & Subjects

Lung Cancer Diagnosis and Treatment Radiomics and Machine Learning in Medical Imaging AI in cancer detection

Publication details

Journal / source
JTCVS Open
Volume
31
Pages
101742–101742
Publication date
30 March 2026
Language
EN