An Optimized Hybrid Framework for Quantitative Segmentation and Progression Analysis of Lung Tumors in CT Imaging

Abstract Accurate segmentation and staging of pulmonary tumors from computed tomography (CT) images are fundamental for early diagnosis, prognosis evaluation, and treatment planning in lung cancer management. This study introduces a Hybrid Spectral–K-Means Clustering Framework that synergistically combines the boundary-preserving characteristics of Spectral Clustering with the centroid refinement capabilities of K-Means, enabling precise delineation of heterogeneous tumor regions. A dataset of 100 Preprocesses lung CT images was Analyzed using a fixed cluster parameter (k = 8) to ensure consistent comparison across unsupervised clustering algorithms, including K-Means, Hierarchical, Gaussian Mixture Model (GMM), Spectral Clustering, and the proposed Hybrid Spectral–K-Means method. Evaluation was performed using standard quantitative metrics: Silhouette Score, Calinski–Harabasz Index, Davies–Bouldin Index, runtime, memory usage, and segmentation accuracy. The Hybrid model achieved superior performance, recording the highest accuracy (95.8%), Silhouette Score (0.5753), and Calinski–Harabasz Index (19,378.03), along with the lowest Davies–Bouldin Index (0.3655). Additionally, the model maintained a low runtime (0.04 s) and minimal memory footprint (800 KB). Tumor staging inferred from segmented tumor area analysis revealed a clinical distribution of Stage I (40%), Stage II (27%), Stage III (13%), and Stage IV (20%). These findings demonstrate that the Hybrid Spectral–K-Means framework offers a robust, interpretable, and computationally efficient unsupervised segmentation approach. It substantially enhances tumor boundary precision and stage estimation accuracy, laying the groundwork for AI-assisted diagnostic systems and radiomics-based lung cancer analysis. Keywords: Lung Tumor Segmentation; Hybrid Clustering; Unsupervised Learning; Computed Tomography (CT); Tumor Staging; Spectral–K-Means; Image Analysis; Radiomics; AI-Assisted Diagnosis.