In this study, we investigate deep learning models for an effective and efficient lesion segmentation in 3D brain diffusion weighted images (DWI). For image segmentation, convolution neural networks (CNN) and Transformer-based models are widely used. ...
In this study, we investigate deep learning models for an effective and efficient lesion segmentation in 3D brain diffusion weighted images (DWI). For image segmentation, convolution neural networks (CNN) and Transformer-based models are widely used. CNNs excel at extracting local features, while Transformer-based models excel at extracting global features. Herein, we employ 2 CNN models (3D-Unet and 3D-UNet++) and 2 Transformerbased models (3D-MobileViT and 3D-SwinUNetR) for brain lesion segmentation. To evaluate the four models, DWI and ADC (apparent diffusion coefficient) of 651 brain stroke patients are used as train, validation, and test set in this study. The experimental results demonstrate that deep learning models are able to successfully segment stroke lesion but their performance varies depending on the size and frequency of the lesion among patients.