An intelligent approach for lung cancer detection and classification

Abstract

Lung cancer remains one of the leading causes of cancer-related deaths worldwide, primarily because it is often diagnosed at an advanced stage when treatment options are less effective. In many cases, early-stage lung cancer is asymptomatic, meaning that patients do not experience noticeable symp- toms until the disease has progressed. This delay in symptom onset leads to a higher likelihood of late detection, by which point tumors may have spread beyond the lungs, significantly reducing the effectiveness of available treatments like surgery, radiation, and chemotherapy. Moreover, the subtle and small size of early-stage lung nodules makes them difficult to detect in medical imaging, as they can easily be missed or misinterpreted by radiolo- gists. Therefore, improving early detection methods is critical for increasing survival rates, as early-stage diagnosis allows for more timely intervention, potentially curative treatments, and better overall patient outcomes. This work investigates advanced methodologies for the early detection and classification of lung cancer, leveraging cutting-edge object detection and deep learning techniques. Lung nodules, often considered one of the earliest indicators of lung cancer, are critical to identify accurately to enhance early diagnosis and improve patient outcomes. To address this, we employ the YOLOv7 architecture, renowned for its speed and precision, to detect lung nodules in medical images, with a particular focus on identifying small, subtle nodules that are often challenging to detect using standard imaging techniques. Following the detection phase, we utilize transfer learning with the pre- trained VGG16 model for the classification of the detected nodules. Transfer learning allows us to harness the power of VGG16’s deep feature extraction while fine-tuning the model to be highly sensitive to lung nodule characteris- tics. Together, these techniques enhance the accuracy, reliability, and speed of nodule detection and classification, addressing the significant challenge of early-stage lung cancer identification. By integrating these models, our ap- proach not only improves diagnostic precision but also serves as a valuable tool for radiologists, aiding them in making more informed, data-driven de- cisions in clinical practice.