MRI诊断140例腰椎智能网络自动检测分型MCs方法的比较

Comparison of MRI diagnosis of 140 cases of MCs using intelligent network automatic detection and classification methods

目的探究使用Yolov5网络检测分类Modic改变(MCs)的性能,与基于Yolov5和Resnet34网络自动检测分类MCs方法进行比较。方法回顾性分析2020年6月至2021年6月接受MRI诊断且符合纳入和排除标准的MCs患者140例,其中男55例,女85例,25~89岁,平均(59.0±13.7)岁。在完成MRI影像的标注工作后,将标注后的MRI影像导入深度学习模型训练,使用医学数据常规增强和Mosaic数据增强进行数据扩充,降低训练数据集过少的因素;利用迁移学习的方法,解决网络在小数据集上过拟合的问题。采用平均精度(AP)、平均精度均值(mAP)、召回率、精确率、F1值等性能指标对两种方法诊断MCs进行评估并比较。结果Yolov5网络检测分类MCs时,mAP、召回率、精确率和F1值分别达到87.56%、82.05%、89.44%和0.845;Yolov5和Resnet34网络自动检测分类MCs时,召回率、精确率和F1值分别达到88.41%、88.68%和0.885。结论Yolov5网络可以帮助诊断腰椎MCs,使用Yolov5和Resnet34网络检测分类MCs时,模型诊断MCs的性能提升,进而表明Yolo系列网络可以为智能辅助诊断技术在脊柱领域的应用提供可能性。

Objective To explore the performance of Yolov5 network to detect and classify Modic changes(MCs)and to comparethe performance of Yolov5 alone and Yolov5 plus Resnet34.Methods Clinical data of 140 patients with MCs who under went MRI diagnosis and met the inclusion and exclusion criteria during June 2020 and June 2021 were retrospectively analyzed,including 55 males and 85 females,mean age(59.0±13.7)years(range 25-89 years).After the MRI images were labeled,they were imported into the deep learning model,and were expanded by using the conventional enhancement of medical data and Mosaic data enhancement to reduce the effects caused by two few factors of the training data sets.Transfer learning method was used to solve the problem of network overfitting on small data sets.Average precision(AP),mean average precision(mAP),recall,precision,F1 and other performance indicators were used to evaluate and compare the two methods.Results When Yolov5 was adopted,them AP,recall,precision and F1 were 87.56%,82.05%,89.44%and 0.845,respectively.When Yolov5 and Resnet34 were used in combination,the recall rate,precision rate and F1 reached 88.41%,88.68%and 0.885,respectively.Conclusion The Yolov5 network model can help diagnose lumbar MCs.When Yolov5 and Resnet 34 models are used together,the performance is improved,which indicates that it is possible to use the Yolo series network in the diagnosis of spine diseases.