Classification of Lung Cancer Subtypes In Thoracic CT-Scan Images Using Gray Level Run Length Matrix (GLRLM) and Support Vector Machine (SVM)

Abstract Background Accurate classification of lung cancer subtypes remains a critical challenge in medical imaging because of inter-observer variability and the need for interpretable diagnostic systems. Deep learning achieves high accuracy, but its black-box nature limits clinical interpretability and trust. Materials and Methods This quantitative descriptive study evaluated an interpretable classification system that combines Gray Level Run Length Matrix (GLRLM) texture features with a Support Vector Machine (SVM) classifier. The study analyzed 260 thorax computed tomography (CT) images from The Cancer Imaging Archive (TCIA), comprising four equal classes: normal, adenocarcinoma, squamous cell carcinoma, and large cell carcinoma. Preprocessing included 36×36 pixel region of interest (ROI) cropping and grayscale normalization, and seven GLRLM features were extracted from four orientations. A one-vs-all multi-class SVM was trained on 200 images and tested on 60 hold-out images, with 5-fold cross-validation. Results The system achieved 100% performance on the hold-out test set, which should be interpreted cautiously given the small and controlled dataset. Five-fold cross-validation yielded an average accuracy of 98.5 ± 1.4%. Normal lungs showed a rough texture with heterogeneous gray-level distribution, adenocarcinoma a rough texture with homogeneous distribution, squamous cell carcinoma a smooth texture with homogeneous distribution, and large cell carcinoma a smooth texture with heterogeneous distribution. Conclusion The GLRLM-SVM approach provides accurate and interpretable classification of lung cancer subtypes, with textural features that correspond to known pathological characteristics. It may support clinical decision-making where model explanations are critical.