Hybrid V-Net And Swin Transformer Deep Learning Model For Brain Tumor Segmentation in Low-Quality MRI Scan

Abstract

Brain tumor segmentation from low-quality magnetic resonance imaging (MRI) remains a challenging task due to noise, resolution variation, and low contrast between tumor and healthy tissues. Improving segmentation accuracy is essential to support more precise diagnosis and treatment planning. This study proposes a hybrid deep learning model that integrates V-Net and Swin Transformer architectures for automatic brain tumor segmentation in multimodal MRI images. The MICCAI BraTS 2020 dataset was used, consisting of T1, T1c, T2, and FLAIR sequences with corresponding segmentation labels. The preprocessing pipeline includes resampling, skull stripping, intensity normalization, and data augmentation. V-Net is employed to extract local spatial features from 3D volumetric data, while the Swin Transformer captures global spatial relationships through a self-attention mechanism. Postprocessing procedures such as thresholding, morphological refinement, and false-positive removal are applied to enhance segmentation quality. The proposed hybrid model achieves Dice scores of 0.8635 for Whole Tumor (WT), 0.7179 for Tumor Core (TC), and 0.8073 for Enhancing Tumor (ET), with additional evaluation using precision, recall, and IoU further confirming its effectiveness. These results highlight the model’s potential to improve automated brain tumor segmentation in low-quality MRI images and support its applicability as an efficient AI-assisted diagnostic tool in clinical practice.