Novel Brain Tumor Detection Model Based on YOLO with Multi-module Fusion

To address the challenges of low detection accuracy and insufficient recognition capability in brain tumor detection, this paper proposes a novel multi-module fused small tumor detection model based on YOLO. Firstly, considering the diverse morphology and complex spatial distribution of brain tumors, a Learnable Local Saliency Kernel Module (LLSKM) is introduced to enhance the network’s ability to extract salient local features. Secondly, a high-resolution triple-branch detection head is designed to better align with the scale distribution of brain tumors, replacing the original YOLO head for large-object detection. Furthermore, to improve the model’s spatial position perception, expand the receptive field, and prevent the loss of fine-grained information, the Manhattan Attention Mechanism and a Multi-Scale Convolutional Attention Module (MSCAM) are incorporated. The proposed model demonstrates enhanced attention guidance, multi-scale feature extraction, and salient region modeling, enabling it to more effectively focus on semantically significant areas and spatial-channel relationships. Experimental results on real brain MRI datasets show that the proposed model significantly improves detection accuracy, recall, and mAP@50 for small tumor targets, validating its effectiveness and advancement.