A Channel Fusion Convolutional Neural Network for Lung Nodule Detection

The automatic and precise detection of lung nodules in 3D Computed Tomography (CT) scans is essential for effective lung cancer screening. However, most current methods process CT images slice-by-slice, overlooking inter-slice information. Exploring the inter-slice correlation of 3D CT images is a potential way to enhance the effectiveness of lung nodule detection even further. To deal with this issue, we design a Channel Fusion Convolutional Neural Network (CFCNNet) aimed at detecting pulmonary nodules. Notably, we design a Group Shuffle module (GS module) to fuse the inter-slice feature, which enhances the discrimination and extraction of corresponding information for nodules of varying sizes within the same set of slices. Our ablation study and experimental results on the publicly accessible LUNA16 dataset indicate that the suggested approach significantly improves detection sensitivity, the Competition Performance Metric (CPM) score reaches 89.69%, outperforms other contemporary detection methods.