基于改进YOLOv12算法的MRI图像脑肿瘤检测研究OA

Objective To propose an improved YOLOv12 algorithm based on the C3k2 wavelet transform convolution module(C3k2_WTConv)and the space-to-depth convolution module(SPDConv),so as to enhance the performance of the original YOLOv12 algorithm for MRI image-based brain tumor detection.Methods An YOLOv12-C3k2_WTConv-SPDConv algorithm was built based on the original YOLOv12 algorithm with the C3k2 module from the Backbone and Neck replaced by the C3K2_WTConv module and some standard Conv modules substituted by the SPDConv module.The proposed algorithm was trained and evaluated for performance using the publicly available MRI brain tumor dataset on the Kaggle platform,which was compared with the original YOLOv12 algorithm,the single shot multibox detector(SSD)algorithm and the real-time detection transformer-large variant(RT-DETR-L)algorithm in terms of various performance metrics.Results When used for MRI image-based brain tumor detection,the YOLOv12-C3k2_WTConv-SPDConv algorithm gained advantages over the original YOLOv12 algorithm,which had the mean average precision at intersection over union threshold 0.5(mAP50)being 0.947,the mean average precision at intersection over union threshold 0.5-0.95(mAP50-95)being 0.748,the Params being 2.08M and the inference speed being 90 frames per second;the YOLOv12-C3k2_WTConv-SPDConv algorithm behaved better than the SSD and RT-DETR-L algorithms.Conclusion The YOLOv12-C3k2_WTConv-SPDConv algorithm functions well in MRI image-based brain tumor detection,enhances the original YOLOv12 algorithm in MRI image-based brain tumor detection and contributes to early screening and diagnosis of brain tumors.[Chinese Medical Equipment Journal,2026,47(3):9-17]