Publication

Abstract : Early detection of this disease by CT scan image analysis plays a vital role in boosting patient survival rates with early-stage lung cancer. During this detection process, manual classification of lung nodules is a tiresome and subjective process. Though deep learning-based algorithmic solutions can produce a high level of precision in classifying images of the mentioned class of images, they can only perform classifications in a blackbox paradigm. This proves problematic because interpretation of such binary outputs becomes challenging when they are of paramount importance. As a consequence of such a problem, we propose a novel framework that integrates GoogLeNet with Grad-CAM-based deep learning solutions for the identification of images associated with the LUNA16 class of images. Unlike similar solutions proposed in prior research studies, our proposed framework achieves a high level of precision with 96.0 % accuracy recorded during tests. This level of precision enables our proposed framework to outperform existing CNN-based solutions and ResNet-50-based solutions by 7. 8 % and 1. 7 %, respectively. This framework also exhibits a high p-value of less than 0.001. Furthermore, unlike existing solutions, our proposed framework identifies 73% of the cases where a majority of the class of images concentrate within the regions of interest. As a consequence, our proposed framework becomes a hallmark of a dependable deep learning-based framework that can play a vital role in boosting patient survival rates with early-stage lung cancer. © 2026 IEEE.