Machine learning significantly aids medical research, particularly in anomaly detection and segmentation. Automating tasks improves physician efficiency in early detection and treatment planning. Accurate anomaly segmentation requires precise localization and delineation of abnormalities like tumors and lesions. Training an algorithm/model for anomaly segmentation requires a large quantity and precise annotations of the anomalies, which are vetted by domain experts as ground truth. The precision of annotation and accuracy of model are widely affected by unwanted entities around the region of interest. Preprocessing steps, such as skull removal from MRI images, are crucial for improving model performance. The proposed research focuses on developing an efficient skull region removal technique from MRI images. Various approaches from image processing to deep learning have been explored for skull region removal, which are dependent on huge dataset and precise annotations. With the advent of foundation models, like SAM and SAM-2, user can provide prompts for quicker and more accurate segmentation. While foundation models like SAM-2 have shown signification performance generic images, their performance when it comes to medical domain is less explored. Hence, fine-tuning is essential for optimal performance of SAM-2 model in the medical domain. The proposed approach fine-tunes the SAM-2 model on the NFBS dataset, achieving an IOU score of ~ 89%. The pipeline’s effectiveness is validated on NFBS dataset, demonstrating its potential as a valuable preprocessing step for MRI image analysis. The current approach is compared with existing approaches publicly available and is found to be in par with the same. While the proposed research work performs well on most of the slices in the MRI data, the boundary slices have scope for improvement. Also rotated images have lesser IOU score as compared to non-rotated images.