MMAP-01 MULTI-MODAL MRI BASED SEGMENTATION OF BRAIN METASTASES USING ADAPTIVE SELF-ATTENTION

Abstract Brain metastases (BMs) are the most common adult central nervous system malignancy, affecting 20–40% of cancer patients. Accurate segmentation of metastatic lesions in multi-modal MRI is essential for treatment planning and prognosis however, manual delineation is time consuming and prone to variability. Traditional deep learning models such as U-Net, have improved segmentation accuracy but capture limited long-range dependencies and struggle with variations in metastasis size, shape, and distribution. This study introduces the Adaptive Integrated Multi-modal Segmentation (AIMS) model, an adaptive self-attention framework within a hybrid U-Net and Transformer architecture to enhance BM segmentation by leveraging multi-modal MRI integration. The proposed model integrates convolutional feature extraction with self-attention to capture both local and global contextual information while filtering out non-informative slices. The BraTS-METS dataset, consisting of 1303 cases with T1, T1Gd, T2, and FLAIR sequences, was used for training and evaluation. Preprocessing included bias field correction, intensity normalization, spatial resampling, and skull stripping. The encoder employs a U-Net backbone, while transformer based attention in the bottleneck refines feature interactions. Feature-wise attention maps guide the decoder to enhance segmentation accuracy, particularly for small and irregularly shaped metastases. The model was validated using a fivefold cross-validation approach and demonstrated superior segmentation performance, achieving higher Dice Similarity Coefficients (DSC) compared to state-of-the-art hybrid U-Net based models. Hausdorff Distance 95 (HD95) scores further indicated precise boundary delineation. By integrating adaptive self-attention with multi-modal MRI, the proposed model enhances segmentation accuracy and robustness in brain metastases. The findings highlight its potential for improving automated BM delineation, reducing manual inefficiencies, and assisting medical professionals in treatment planning and informed clinical decisions.