Reliable segmentation of brain tumors in intraoperative and noisy MRI settings continues to present substantial difficulties arising from magnetic field inhomogeneity, tumor microstructural variation, and elevated image noise. This work introduces a synergistic method that combines the Segment Anything Model (SAM) with the YOLO object detector and convolutional neural network (CNN) architectures to elevate both the precision and the resilience of tumor delineation in real-time surgical environments. The framework capitalizes on SAM’s zero-shot and user-guided segmentation, YOLO’s robust spatial localization, and CNN’s deep feature extraction, collectively mitigating artifacts from non-uniform intensity and intralesional architectural diversity. Comprehensive validation on both the BRATS 2020 database and intraoperative MRI collections indicates marked enhancement in Dice Similarity Coefficient and Intersection over Union metrics—especially in imaging marred by excess noise. This integrated system thus promises rapid clinical utility in diagnostic triage, precise treatment sculpting, and ongoing therapeutic assessment, advancing the frontiers of AI-augmented medical imaging.