Segmentation of Cerebral Hemorrhage CT Images using Swin Transformer and HarDNet
Zhegao Piao, Yeong Hyeon Gu, Seong Joon Yoo, Myoungho Seong · 2023
Segmentation is a technique that divides an image by relevancy into pixel units and is often used to detect diseases in medical imaging analysis. Furthermore, MRI and CT are distinct data types commonly used in medical imaging analysis. The difference is that MRI analyzes 3D images, whereas CT involves 2D images. In other words, it is not certain that a model with a superb performance in MRI images will also be ideal in CT images, as it relates to the generality of the data and the model. In this study, we combined HarDNet and Swin Transformer to describe methods and experimental processes that improve performance compared to using the two individual models separately. In addition, STHarDNet, a combination of HarDNet and Swin Transformer, which performed well in MRI images, was applied to cerebral hemorrhage CT images provided by AIHub. The experiment results showed that STHarDNet exhibited superior performance in cerebral hemorrhage CT images in addition to MRI images compared to seven other models, including U-Net, U-Net++, SegNet, PSPNet, Swin UNet, HarDNet, and Swin Transformer.