A Hybrid Multimodal Framework for Lung Cancer Segmentation and Classification Using Fuzzy–Possibilistic Clustering

Abstract

Lung cancer remains a major global health concern, highlighting the need for accurate and reliable computer-aided diagnostic systems for medical image analysis. The segmentation of pulmonary nodules and the classification of suspicious lung lesions are challenging because of variations in nodule morphology, intensity distribution, anatomical structures, and image noise. This study proposes a hybrid multimodal framework for lung cancer segmentation and classification using fuzzy–possibilistic clustering and integrated image-processing techniques. The proposed framework comprises image acquisition, preprocessing, lung region extraction, multimodal feature integration, fuzzy–possibilistic clustering, region refinement, feature extraction, and classification. Medical images are standardized through intensity normalization, spatial alignment when required, and noise reduction to improve image quality and feature consistency. Fuzzy–possibilistic clustering combines fuzzy membership information with possibilistic typicality estimation to support the identification of heterogeneous lesion regions while reducing sensitivity to ambiguous pixels and noise. Texture, intensity, and morphological features are subsequently extracted from the segmented regions and used to train a supervised classification model for distinguishing suspicious lesions from non-suspicious regions or classifying available diagnostic categories. The framework is designed to be evaluated using segmentation metrics, including the Dice similarity coefficient and intersection over union, together with classification measures such as accuracy, sensitivity, specificity, precision, F1-score, and receiver operating characteristic area under the curve. Comparative experiments with conventional clustering and segmentation techniques can establish the effectiveness and limitations of the proposed approach.