MLS-YOLOv8m: A Multi-Scale Feature Enhancement Method for Pulmonary Nodule Detection

Aiming at the problem that the convolutional neural network-based pulmonary nodule detection method is prone to lose detailed features in the deep-layer network, a multi-scale feature enhancement method for pulmonary nodule detection(MLS-YOLOv8m) is proposed. Firstly, to enhance the expression of shallow features, the Convolutional Block Attention Module (CBAM) is introduced into the shallow network of the backbone network to retain more detailed information; Secondly, to reduce the information loss of deep features during the pooling process, SoftPool operation is used instead of the Max-pooling operation in Spatial Pyramid Pooling - Fast (SPPF), which is beneficial for the aggregation of multi-scale features; Finally, for extracting more refined multi-scale features, a new module (ResC2f) is designed to replace the original C2f module, thereby improving the performance of the model. The proposed method is utilized for detecting pulmonary nodules in CT images from the LUNA16 dataset, demonstrating its capacity to reduce model parameters while maintaining strong detection performance.