An improved attention mechanism based YOLOv4 structure for lung nodule detection

Abstract

In order to detect early lung cancer and improve the detection accuracy of pulmonary nodules detection in CT images, we propose an improved attention mechanism-based YOLOv4 lung nodules detection method. The detection steps are as follows: Data is resampled and preprocess by lung parenchyma segmentation. Improve the YOLOv4 network structure, add the empty convolution to increase the receptive field of shallow feature maps, and integrate the attention mechanism to enhance useful feature capture. Advanced model was trained using the sub-dataset LUNA16 of LIDC/IDRI and evaluated using the test set. The final evaluated model has good robustness, and the accuracy can reach 89.11%. The experimental results show that our practice can effectively detect lung nodules.