Automatic pulmonary nodule detection on computed tomography images using novel deep learning

Lung cancer poses a significant threat, contributing significantly to cancer-related mortality. Computer-aided detection plays a pivotal role, particularly in the automated identification of pulmonary nodules, assisting radiologists in diagnosis. Despite the remarkable efficacy of deep convolutional neural networks in lesion identification, the detection of small nodules remains an enduring challenge. A conventional automated detection framework encompasses two critical stages: candidate detection and false positive reduction. This study introduces a novel approach named ReRointNet, focusing on meticulous lung nodule localization and detection through strategically placed sample points. To enhance nodule detection, we propose integrating PointNet anchors with RPN anchors. PointNet, operating on local key points, facilitates this integration. The synergy achieved by merging these anchors within our RePointNet framework enhances nodule detection rates and substantially improves localization accuracy. Post-detection, identified nodules undergo classification using the 3D Convolutional Neural Networks (CNN) method. Our contribution presents a novel paradigm for nodule detection in lung Computed Tomography (CT) images, with reduced computational costs and improved memory efficiency. The combined utilization of RePointNet and 3DCNN demonstrates proficiency in identifying nodules of various sizes, including small nodules. Our research underscores the superiority of lung nodule identification through the utilization of RePointNet based on point information, surpassing conventional networks. Rigorous evaluations of the LUNA16 dataset reveal our method's superior performance compared to state-of-the-art systems, achieving a notable sensitivity of 91.6 percent at a speed of 0.9 frames per second. These findings underscore the potential of our proposed approach in advancing precise lung nodule diagnosis, offering invaluable support to healthcare practitioners and radiologists engaged in diagnosing lung cancer patients.