DETECTING LUNG CANCER WITH DEEP CNN MODELS

Abstract

Due to Lung cancer, it is one of the most life-threatening diseases worldwide, and its early detection is essential for improving survival outcomes and reducing cancer-related mortality. Conventional diagnostic procedures based on computed tomography images and chest radiographs largely depend on expert interpretation, which may be affected by inter-observer variability, workload pressure, subtle lesion appearance, and delayed clinical decision-making. Recent advances in deep learning, particularly Convolutional Neural Networks, have demonstrated strong potential in automated medical image analysis by learning complex spatial and hierarchical features directly from imaging data. This research proposes an explainable deep CNN-based framework for lung cancer detection using medical imaging data, with emphasis on accurate classification, robust feature extraction, and clinical interpretability. The proposed approach includes image preprocessing, noise reduction, normalization, lung region enhancement, data augmentation, transfer learning, and CNN-based classification to distinguish cancerous and non-cancerous lung images. Multiple deep learning architectures, including custom CNN, VGG16, ResNet50, DenseNet121, InceptionV3, and EfficientNet, are considered for comparative evaluation to identify the most effective model for lung cancer diagnosis. Performance is assessed using accuracy, precision, recall, specificity, F1-score, receiver operating characteristic analysis, and confusion matrix-based evaluation. To improve transparency, Gradient-weighted Class Activation Mapping is integrated to highlight suspicious lung regions that influence the model’s prediction. The study aims to reduce false positive and false negative predictions while improving diagnostic reliability in early-stage lung cancer detection. The proposed framework can assist radiologists as a decision-support tool by providing faster, consistent, and explainable predictions from lung imaging data. The research contributes toward the development of intelligent, interpretable, and clinically supportive computer-aided diagnosis systems for lung cancer detection.