| Abstrakt: |
Brain tumors, characterized by the growth of aberrant and unregulated cells within the skull, pose a significant threat to normal brain function and, if malignant, can present life-threatening consequences. Timely identification and diagnosis through computed tomography (CT) or magnetic resonance imaging (MRI) are pivotal for effective intervention. This study introduces a streamlined image-processing strategy aimed at precisely localizing and identifying brain tumor regions. The methodology involves successive phases including preprocessing, edge detection, and segmentation. The initial phase converts the original image into a grayscale format while incorporating essential noise-reduction techniques. Subsequently, image enhancement methods are applied to improve the image quality. The pivotal step of segmentation employs a UNET trained on meticulously defined tumor mask regions, facilitating accurate delineation of tumor boundaries. The proposed methodology demonstrates exceptional performance metrics, achieving an impressive accuracy of 99.67%. Furthermore, it attains notably high specificity and precision scores of 99.73% and 98.79%, respectively. The sensitivity metric also demonstrates robust performance at 97.43%. This comprehensive approach offers a promising avenue for precise and automated brain tumor detection and classification in medical imaging, providing clinicians with a reliable tool for early and accurate diagnosis, critical for timely intervention and patient care. |
Full text from SpringerLink