Interpretable MedDeepNet: Deep feature learning with atom search optimization for explainable lung cancer detection in CT images

Lung cancer remains one of the leading causes of cancer-related mortality worldwide, with early and accurate detection being crucial for improving survival rates. Deep learning techniques (DL) have recently demonstrated significant potential in the automated analysis of computed tomography (CT) scans for medical diagnosis. However, existing approaches typically involve a large number of parameters, substantial computational cost, and limited interpretability, which pose challenges for both transparent decision-making and stable training, especially when data is limited. Therefore, this study proposes an interpretable MedDeepNet framework to enhance lung cancer detection and provide clinically meaningful explanations of the model’s decisions. The proposed model is evaluated on three publicly available datasets, namely IQ-OTH/NCCD, LIDC-IDRI, and lung CT scan dataset. The proposed architecture employs a deep feature extraction network to capture both local texture patterns and high-level structural information from chest CT images. To improve classification efficiency, Atom Search Optimization (ASO) is utilized for feature selection, reducing the discriminative features, which are subsequently classified using SVM and KNN classifiers. Furthermore, an adaptive superpixel perturbation-based Local Interpretable Model-Agnostic Explanation (LIME) framework is employed to enhance the explainability of the proposed model by generating clear and interpretable visual explanations that highlight the most influential image regions. The proposed method achieved an accuracy of 99.84%, with a recall of 99.79%, precision of 99.87%, specificity of 99.93%, and an F1-score of 99.83% on the IQ-OTH/NCCD dataset. In addition, the model produces informative visualizations that reveal the critical regions in CT scans influencing its diagnostic decisions.