INTEGRATIVE APPROACHES FOR DETECTING LUNG CANCER USING DEEP LEARNING ALGORITHMS

Abstract

Among cancer-related deaths worldwide, lung cancer remains one of the leading causes, primarily due to late diagnosis and limited treatment options at advanced stages. Early detection through medical imaging plays a crucial role in improving survival rates; however, many existing automated detection methods rely on a single imaging modality and limited datasets, reducing their reliability in clinical settings. This paper presents a dual-modality, multi-hybrid deep learning approach for lung cancer detection using chest X-ray and computed tomography (CT) images. Each imaging modality is trained separately on carefully prepared and augmented datasets using transfer learning to enhance robustness and generalization. MobileNet and DenseNet architectures are employed to effectively capture modality-specific features. The chest X-ray dataset is categorized into three classes, while the CT scan dataset consists of two classes, reflecting their diagnostic differences. A combined data generator is used to train the dual-modality framework, enabling effective learning from both image sources. Experimental results show that the CT-based model achieves an accuracy of 93%, while the X-ray-based model attains an accuracy of 89.50%. These findings indicate that modality-specific learning within a dual-modality framework improves detection performance and supports more reliable AI-assisted lung cancer diagnosis.