Brain Tumor MRI Segmentation Method Based on Segment Anything Model

Abstract

The precise segmentation of different types of brain tumor regions constitutes a critical task in medical image segmentation.Clinically, brain MRI contains abundant information, which can significantly assist doctors in the examination and diagnosis of brain tumor patients.With the advancement of artificial intelligence (AI) and computer technology, some foundational models have increasingly played a pivotal role in the field of computer vision.The Segment Anything Model (SAM) is a fundamental model in the realm of image segmentation, renowned for its exceptional zero-shot segmentation performance and transfer ability, achieving commendable results in natural image processing.To explore the efficacy of SAM in segmenting brain tumor MRI and address the issue of low segmentation accuracy due to uneven image grayscale, a method based on SAM feature fusion is proposed.Features fused from the Transformer and Convolutional Neural Network (CNN) are input into a mask decoder, leveraging the attention mechanism of the Transformer to more effectively capture the global relationships within images, thereby enhancing the precision of the output.Experiments have demonstrated that the method proposed in this study surpasses the segmentation performance of SAM alone, achieving precise segmentation of brain tumor MRI.