Segmentation of Pulmonary Nodules Based on Improved UNet++

Lung cancer has the highest incidence and mortality rates worldwide. Lung nodule segmentation of the computed tomography (CT) images plays a crucial role in early lung cancer diagnosis. This paper proposes a CT image lung nodule segmentation method based on ResNeXt and UNet++ combined with the SCSE attention module. This network has strong pulmonary nodules feature extraction capabilities. It enables the network to express pulmonary nodules features with finer granularity while increasing the receptive field of each layer of the network. The image is preprocessed by k-means clustering to extract the ROI image of the lungs in CT scan slices, which can reduce the search space of the model. The evaluation of the suggested model was performed by utilizing the LIDC-IDRI dataset. According to the results, we found that our network showed good performance, achieving an 83.43% Dice score value and 72.51% IOU score.