BiTr-Unet: Integrating CNN and Transformer Architectures for Accurate Brain Tumor Segmentation

BiTr-Unet: Integrating CNN and Transformer Architectures for Accurate Brain Tumor Segmentation

Kia Shirdel1

1) Department of Computer Engineering from Shahid Mehdi Bakri Faculty of Technology and Engineering, Islamic Azad University, Urmia, Iran

IOI Code : XKBA-AAZZFC
Publication : 3rd International Conference on Recent Advances in Engineering, Innovation and Technology - Belgium(eitconf.com/3rd)
Abstract :
    Brain tumor segmentation plays a critical role in medical imaging, enabling precise diagnosis and treatment planning. This paper introduces a novel framework that integrates Convolutional Neural Networks (CNNs) and Transformer architectures to enhance segmentation accuracy and robustness. The proposed model leverages CNNs for extracting local spatial features and Transformers for capturing long-range dependencies, addressing limitations in existing single-architecture approaches. Using the BraTS 2023 dataset, the model achieved state-of-the-art performance with a Dice Score of 0.91 and improved generalization across diverse tumor types. Comprehensive experiments demonstrate the superiority of our approach compared to baseline methods, highlighting its potential for clinical applications.

Keywords : brain tumor segmentation convolutional neural networks segmentation accuracy dice score transformer architectures real time deployment