Over the past fifteen years, deep learning has revolutionized medical imaging, particularly in the automated detection of brain tumors. This paper presents a domain-adapted object detection framework, YOLOv8-Med, specifically optimized for clinical applications. Building upon advances in Convolutional Neural Networks (CNNs), the model integrates depthwise separable convolutions, attention-driven modules, and medical-domain-specific feature extractors to enhance detection accuracy without compromising speed. The proposed system was benchmarked against established architectures including YOLOv8x, YOLOv7, EfficientDet-D7, and Faster R-CNN. Experimental evaluations on standard datasets demonstrated the superiority of YOLOv8-Med, achieving a precision of 93.5%, mAP@0.5 of 94.8%, and an inference speed of 49 FPS. These metrics affirm its potential for real-time deployment in clinical settings. Additionally, the model showed improved delineation of tumor boundaries and contextual understanding of complex radiographic features. Future work will investigate integration with multi-modal data (e.g., PET, CT), domain adaptation techniques, and edge deployment for point-of-care diagnostics. This study underscores the potential of optimized CNN-based frameworks to augment radiological decision-making and accelerate the adoption of AI in medical workflows.

Keywords: Brain Tumor Detection, Medical Imaging, YOLOv8-Med, Convolutional Neural Networks (CNN), Real-time Object Detection, Multi-scale Feature Fusion, Edge Deployment