eng Oriental Scientific Publishing Company Biomedical and Pharmacology Journal 0974-6242 2026-03-20 19 1 625 641 10.13005/bpj/3380 70429 article Pooja Dwivedi 1 Sagar Barage 2 Ashish Kumar Jha 3 Archi Agrawal 1,3 Venkatesh Rangarajan 1,3 Department of Nuclear Medicine and Molecular Imaging, Advanced Centre for Treatment Research and Education in Cancer, Tata Memorial Centre, Homi Bhabha National Institute, Mumbai, India, Amity Institute of Biotechnology, Amity University Maharashtra, Mumbai-Pune Expressway, Bhatan, Panvel-410206, India Department of Nuclear Medicine and Molecular Imaging, Tata Memorial Hospital, Tata Memorial Centre, Homi Bhabha National Institute, Mumbai, India Independent validation is critical in radiomic research to mitigate bias and enhance clinical translatability. This study presents the development and external validation of a PET (Positron Emission Tomography) radiomics-based machine learning (ML) model for the non-invasive classification of non-small cell lung cancer (NSCLC) histological subtypes, adenocarcinoma and squamous cell carcinoma, and evaluates its clinical utility. Adhering to the CLEAR radiomics reporting guidelines, two retrospective PET datasets from different institutions were employed: one for model training and the other for external validation. Tumor segmentation was performed using a 40% SUVmax threshold, followed by radiomic feature extraction using IBSI-compliant software and ComBat harmonization. To address class imbalance, appropriate resampling techniques were applied, and recursive feature elimination identified the top five predictive features. Three ML models were trained and evaluated on unseen external data. Model performance was compared using Delong’s test, and clinical utility was assessed via decision and clinical impact curve analyses. The XGB model demonstrated superior performance with an accuracy of 80% and an AUC of 0.82 [95% CI: 0.73–0.91], along with favourable calibration (Brier score: 0.12). Decision curve analysis confirmed its net clinical benefit. These findings underscore the clinical relevance of PET radiomics combined with advanced ML techniques, offering a robust, non-invasive tool for histological subtype classification in NSCLC. The externally validated model demonstrates strong potential for integration into translational oncological workflows, aiding personalized treatment planning and improving diagnostic precision in real-world settings. https://biomedpharmajournal.org/vol19no1/external-validation-of-pet-radiomics-based-ml-model-for-non-invasive-lung-cancer-subtype-classification/ External Validation Histological Subtypes Lung Cancer Machine Learning PET Radiomics