RepSE-CBAMNet: A Hybrid Attention-Enhanced CNN for Brain Tumor Detection

The effective detection of brain tumors is closely linked to their timely diagnosis and treatment which can help in the prevention of deaths and in improving the quality of life. The objective of this paper is to present an enhanced YOLO (You Look Only Once) architecture designed specifically for brain tumor detection, significantly improving detection accuracy using Convolutional Block Attention Modules (CBAMs), Squeeze-and-Excitation Blocks (SE), and Residual Blocks. The use of CBAM enhances the model's ability to focus on critical spatial and channel-wise features in complex medical images. RepVGG blocks offer efficient feature extraction while Residual blocks help mitigate the vanishing gradient problem. Squeeze-and-Excite (SE) blocks further improve feature representation by emphasizing important channels. Evaluated on three brain tumor datasets and compared to RepVGG-GELAN, YOLOv9c, RCS-YOLO and Yolov5L, our model demonstrates significant improvements, in precision and AP50:95, for the first dataset and competes very closely with them. This architecture offers a promising approach to improving the detection of brain tumors, supporting more accurate diagnoses and potentially enhancing clinical decision-making. The implementation code is publicly available at https://github.com/GKatsagannis/RepSE-CBAMNet.