Medical SAM: A Glioma Segmentation Fine-tuning Method for SAM Using Brain MR Images.

Based on the 2016 World Health Organization (WHO) Classification scheme for gliomas, accurate glioma segmentation serves as a fundamental basis for glioma diagnosis. In the realm of glioma diagnosis, brain MRI has emerged as an indispensable diagnostic tool due to its ability to provide comprehensive information. Over the past decade, there has been a notable surge in the utilization of machine learning techniques, particularly deep learning, for processing medical images. These deep learning methods, based on convolutional neural networks or transformers, proposed analogous architectures such as U-Net for precise segmentation of medical images, thereby significantly enhancing the accuracy of brain glioma segmentation. Thanks to the development of foundation models, models pre-trained with large-scale datasets have achieved better results on a variety of tasks. However, for medical images with small dataset sizes, deep learning methods struggle to achieve better results than on real-world image datasets. In this study, we proposed an adapter to effectively fine-tune the foundation model (SAM) for improved glioma segmentation using brain MR images. The effectiveness of the proposed method is validated via our private glioma data set from the First Affiliated Hospital of Zhengzhou University (FHZU) in Zhengzhou, China. The proposed method outperforms current state-of-the-art methods, achieving a Dice coefficient of 87.33% and a Hausdorff distance of 10.87 for accurate segmentation of the glioma region in glioma treatment, representing a significant 4% improvement in Dice coefficient.