摘要
为应对磁共振成像(MRI)中形态复杂和边界不规则的脑肿瘤检测,提出一种改进的YOLOv8模型TumorNet-YOLO。该模型通过三项创新模块提升检测性能:自适应感受野卷积模块增强多尺度肿瘤特征的提取能力,降低漏检率;分割融合卷积模块通过多尺度特征融合,增强浅层和深层特征的协同作用;可变形融合模块优化对不规则肿瘤区域的检测,提升模型在复杂MRI背景下的鲁棒性。实验结果表明,TumorNet-YOLO在脑肿瘤检测数据集Br35H中表现优异,平均精度均值mAP@0.5为96.6%,mAP@0.5∶0.95为73.8%。此外,模型计算量(GFLOPs)为8.6,显著优于现有方法。为了验证模型的泛化能力,在BCCD和BTOD数据集上进行了对比实验,结果显示TumorNet-YOLO在mAP@0.5和mAP@0.5∶0.95等多个指标上超越了YOLOv8n,表明TumorNet-YOLO可为脑肿瘤检测和医学图像分析提供更为有效的解决方案。
In order to address the challenges of detecting brain tumors with complex shapes and ir-regular boundaries in Magnetic Resonance Imaging(MRI),an improved YOLOv8 model,Tumor-Net-YOLO,is proposed.This model enhances detection performance through three innovative mod-ules:the Adaptive Receptive Field Convolution Module improves the ability to extract multi-scale tumor features and reduce false negatives;the Segmentation Fusion Convolution Module strengthens the synergy between shallow and deep features through multi-scale feature fusion;the Deformable Fusion Module optimizes the detection of irregular tumor regions and improves the model’s robust-ness in complex MRI backgrounds.Experimental results show that TumorNet-YOLO performs ex-cellently on the brain tumor detection dataset Br35H,with a mean Average Precision mAP@0.5 of 96.6%and mAP@0.5∶0.95 of 73.8%.Furthermore,the model’s computational cost is 8.6 GFLOPs,which significantly outperforms existing methods.To validate the model's generalization a-bility,comparative experiments were conducted on the BCCD and BTOD datasets.The results dem-onstrate that TumorNet-YOLO outperforms YOLOv8n in multiple metrics,including mAP@0.5 and mAP@0.5∶0.95,indicating that TumorNet-YOLO provides an effective solution for brain tumor detection and medical image analysis.
作者
张傲
刘微
刘阳
杨思瑶
管勇
李波
刘芳菲
ZHANG Ao;LIU Wei;LIU Yang;YANG Siyao;GUAN Yong;LI Bo;LIU Fangfei(Shenyang Ligong University,Shenyang 110159,China)
出处
《沈阳理工大学学报》
2025年第5期8-13,20,共7页
Journal of Shenyang Ligong University
基金
辽宁省教育厅高等学校基本科研项目(JYTMS20230189)
沈阳理工大学引进高层次人才科研支持计划项目(1010147001131)。
