<![CDATA[Brain tumors are abnormal tissue growths in the brain. These brain tumors can have a negative impact on human health, one of which can interfere with brain functions such as vision, balance, and so on. Therefore, early detection needs to be done, one of which is by using medical imaging modalities, i.e., MRI. However, analyzing MRI scans requires careful observation and a high level of proficiency. Thus, medical image segmentation is required. Segmentation is important in medical image analysis as it allows medical experts to distinguish between abnormal and normal tissues. This study aims to determine the ability of the swin transformer architecture in segmenting brain tumor MR images. The image data used was BraTS 2021 data with a total of 1,250 images. The data were divided into three, i.e., training set, validation set, and testing set with a ratio of 70:15:15. Swin Transformer provided two main concepts, i.e., hierarchical feature maps and attention window shifts. The Swin Transformer initially was divided the image into small patches, which were then converted into vector form. After that, it was passed through W-MSA for local area and SW-MSA for cross window area. Next, multiple patches were merged into one, so that the image resolution gradually decreased, and then restored back to the original resolution. Based on this, the segmentation results were evaluated using a confusion matrix using DSC, IoU, and sensitivity metrics. The results of brain tumors MR image segmentation with Swin Transformer obtained evaluation values, i.e., 0.97313 for DSC, 0.94767 for IoU, and 0.96450 for sensitivity. It can be concluded that the Swin Tranformer can effectively segment brain tumor MR images.]]>
