基于CNN的CT切片骨密度定性检测用于骨质疏松筛查
本研究利用卷积神经网络(CNN)分析腰椎L1椎弓根水平的诊断性2D CT切片,构建MS-Net分割ROI和BMDC-Net分类骨量,在150例患者上训练、63例患者上测试,BMDC-Net准确率76.65%,AUC为0.9167,提示该方法有望用于临床骨质疏松筛查。
要点
- MS-Net在腰椎不同形态保持上分割精度优异,Dice指数、像素准确率和IOU均大于0.8。
- BMDC-Net对正常骨量、低骨量和骨质疏松的分类准确率为76.65%,AUC为0.9167。
- 该方法可融入常规CT诊断流程,有望减少放射科医生手工负担和患者诊断成本。
核心结论
- 基于CNN分析诊断性CT切片定性检测骨密度的方法在骨质疏松筛查中具有临床应用潜力,可减少放射科医生手工负担和患者诊断成本。
研究设计与样本
- 研究设计为利用CNN对诊断性2D CT切片进行骨密度定性检测,方法包含MS-Net定位分割ROI和BMDC-Net分类骨量两个功能模块。
- 实验选取腰椎L1椎弓根水平的诊断性2D CT切片,网络在150例患者上训练,在63例患者上测试,每位患者对应一张2D CT切片。
主要发现
- MS-Net在不同腰椎形态保持上分割精度优异,分割结果的Dice指数、像素准确率和IOU均大于0.8。
- BMDC-Net获得正常骨量、低骨量和骨质疏松的概率值,准确率为76.65%,受试者工作特征曲线下面积(AUC)为0.9167。
- 放射科医生可基于网络结果提供骨密度的初步定性诊断结果。
临床怎么用
- 可将该方法融入常规CT诊断流程,在体检中利用已有诊断性CT切片进行骨质疏松机会性筛查。
- 放射科医生可参考网络输出的正常骨量、低骨量和骨质疏松概率值,提供初步定性诊断,减少手工负担。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
UNLABELLED: The features extracted from diagnostic computed tomography (CT) slices were used to qualitatively detect bone mineral density (BMD) through neural network models, and the evaluation results indicated that it may be a promising approach to perform osteoporosis screening in clinical practice. INTRODUCTION: The purpose of this study is to design a novelty diagnostic method for osteoporosis screening by using the convolutional neural network (CNN), which can be incorporated into the procedure of routine CT diagnostic in medical examination thereby improving the osteoporosis diagnosis and reducing the patient burden. METHODS: The proposed CNN-based method mainly comprises two functional modules to perform qualitative detection of BMD by analyzing the diagnostic 2D CT slice. The first functional module aims to locate and segment the ROI of diagnostic 2D CT slice, called Mark-Segmentation-Network (MS-Net). The second functional module is used to determine the category of BMD by the features of ROI, called BMD-Classification-Network (BMDC-Net). The diagnostic 2D CT slice of pedicle level in lumbar vertebrae (L1) was selected from 3D CT image in our experiments firstly. Then, the trained MS-Net can get the mark image of input original 2D CT slice, thereby obtain the segmentation image. Finally, the trained BMDC-Net can obtain the probability value of normal bone mass, low bone mass, and osteoporosis by inputting the segmentation image. On the basis of network results, the radiologists can provide preliminary qualitative diagnosis results of BMD. RESULTS: Training of the network was performed on diagnostic 2D CT slices of 150 patients. The network was tested on 63 patients. Each patient corresponds to a 2D CT slice. The proposed MS-Net has an excellent segmentation precision on the shape preservation of different lumbar vertebra. The dice index (DI), pixel accuracy (PA), and intersection over union (IOU) of segmentation results are greater than 0.8. The proposed BMDC-Net achieved an accuracy of 76.65% and an area under the receiver operating characteristic curve of 0.9167. CONCLUSIONS: This study proposed a novel method for qualitative detection of BMD via diagnostic CT slices and it has great potential in clinical applications for osteoporosis screening. The method can potentially reduce the manual burden to radiologists and diagnostic cost to patients.
引用
- Tang C, Zhang W, Li H, et al. CNN-based qualitative detection of bone mineral density via diagnostic CT slices for osteoporosis screening. Osteoporos Int. 2020. PMID: 33165630. doi:10.1007/s00198-020-05673-w
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。