Combining CNNs and Transformers: YOLOv12-SAM Framework for Brain Tumor Detection and Segmentation

The growth of cells in the brain tissue is an abnormal growth, which may be benign or malignant. Early diagnosis and accurate segmentation of brain tumors are critical for diagnosis, treatment planning and outcome of patient. Recently Computer vision, particularly deep learning algorithms, plays a vital role in brain tumor analysis. In this study, we propose the usage of two deep learning models for brain tumor detection and segmentation; namely YOLOv12 for detection and Segment Anything Model (SAM) for segmentation. The tumor is first detected by YOLOv12with a bounding box, which is then used as a prompt for the Segment Anything Model (SAM) to accurately segment the tumor region. The speed and accuracy of YOLOv12 are used in the first part of the proposed method to detect potential tumor regions in brain MRI scans, with SAM then used to precisely delineate tumor boundaries detected by YOLOv12.With the help of the combined YOLOv12-SAM model approach achieves high segmentation accuracy 99.79% and Dice similarity coefficient 93.23% comparable with state-of- art models. This integrated framework provides a robust, efficient and very accurate automatic analysis of brain tumors in the clinical environment and may substantially enhance the diagnostic capabilities and treatment planning.