Year: 2026 | Month: June | Volume 13 | Issue 1

Grad-CAM and LIME-based Interpretations of Deep Ensemble Strategy for Lung Cancer Classification

Pragnya Das1* Satya Narayan Tripathy2 Sunil Kumar Pradhan3 and Kali Prasad Rath4
DOI:10.30954/2348-7437.1.2026.2

Abstract :

This research presents an integrative framework that utilizes Gradient-weighted Class Activation Mapping (Grad-CAM) and Local Interpretable Model-agnostic Explanations (LIME) for generating reliable and transparent justifications in deep ensemble models tailored for pulmonary carcinoma classification. This research approach used three different pre trained models as ResNet50, DenseNet121, EfficientNetB0 and deep ensemble model. The paper details a methodological pipeline that begins with preprocessing high-resolution computed tomography (CT) images, followed by enhancement and feature extraction using convolutional neural network architectures. As there is no clear suggestion of which pre-trained model gives best performance as compared to others, an ensemble approach has introduced. The ensemble approach consolidates predictions from multiple deep learning models to improve overall accuracy and reduce variance among individual model predictions. To address the critical issue of interpretability in clinical settings, two prominent post-hoc explainability methods— Grad-CAM and LIME—are in use to elucidate the inner workings of the model. The experimental evaluation conducted on CT scan lung cancer imaging datasets shows that the combination of ensemble learning with these XAI techniques enhances both diagnostic performance and transparency. The deep ensemble model achieved 0.972 of accuracy while ResNet50, DenseNet121 and EfficientNetB0 achieved 0.94, 0.9358 and 0.94 of accuracy respectively. The results clearly identify the regions of interest and diagnostic features that contribute to lung cancer detection, thereby increasing clinical trust in the system. The paper also discusses potential challenges, such as the variability in imaging quality and the computational overhead of ensemble approaches, and proposes strategies for optimization and real-time application. Overall, the findings demonstrate that leveraging Grad-CAM and LIME in a deep ensemble framework not only improves predictive accuracy but also provides interpretable insights crucial for clinical decision-making and subsequent patient management.

• PDF Download