Year: 2025
Type: article
Abstract: Accurate segmentation of brain magnetic resonance imaging (MRI) scans is vital for early diagnosis and treatment planning of brain tumors. Classical methods such as the Watershed algorithm often suffer from over-segmentation, noise sensitivity, and limited adaptability. To address these issues, we propose a Watershed-based Multi-Agent System (WAMAS) that combines empirical thresholding, statistical similaritymeasures, and agent-driven negotiation for robust tumor delineation. In preprocessing, edge features are extracted with Canny and Sobel operators, while region descriptors are obtained via Quadtree decomposition and refined through mean–variance analysis to adapt thresholds under noise. During processing, RegionAgents propose the proposed local watershed on its appropriate regions where seed candidate merges based on similarity scores, while Edge Agents validate boundaries using gradient consistency; conflicts are resolved through cooperative decision rules to prevent over-segmentation. Evaluations on BrainWeb and IBSR167 datasets under varying noise levels showed that WAMAS outperforms baseline Watershed andadvanced methods such as U-Net and B-UNet, and best results obtained are respectively 97.38% accuracy, 96.50% sensitivity, and 96.84% specificity. Paired t-tests (p < 0.01) confirmed significant improvements. These results demonstrate that WAMAS provides coherent boundaries and robust performance, making ita promising tool for clinical neuroimaging.
Source: Informatica
Language: English
Subfield: Neurology
Field: Neuroscience
Domain: Life Sciences
Open Access status: diamond
OpenAlex ID: W4417444230