«上一篇/Previous Article|本期目录/Table of Contents|下一篇/Next Article»

《天津医科大学学报》[ISSN:1006-8147/CN:12-1259/R]

卷:

28卷

期数:

2022年02期

页码:

205-210

栏目:

技术与方法

出版日期:

2022-03-20

文章信息/Info

Title:

Preliminary study on CT application of deep convolutional neural network in automatic diagnosis of rib fractures

文章编号:

1006-8147(2022)02-0205-06

作者:

孟祥虹¹; 吴迪嘉²; 马信龙³; 刘爱娥²

（1. 天津市天津医院放射科，天津300211；2. 上海联影智能医疗科技有限公司，上海201210；3. 天津市天津医院骨科，天津300211）

Author(s):

MENG Xiang-hong¹; WU Di-jia²; MA Xin-long³; LIU Ai-e²

（1. Department of Radiology，Tianjin Hospital，Tianjin 300211，China；2. Shanghai United Imaging Intelligence Co.，Ltd.，Shanghai 201210，China； 3. Department of Orthopedics，Tianjin Hospital，Tianjin 300211，China）

关键词:

人工智能; 深度学习; 卷积神经网络; 肋骨骨折; CT

Keywords:

artificial intelligence; deep learning; convolutional neural network; rib fracture; CT

分类号:

R445

DOI:

-

文献标志码:

A

摘要:

目的：探讨深度卷积神经网络（DCNN）模型在胸部CT图像上对肋骨骨折自动定位和诊断的作用。方法：回顾性纳入2 300例因胸外伤于门急诊初诊、行胸部CT检查的患者图像，其中300例为测试集。应用分割网络、关键点检测网络和骨折检测网络建立DCNN模型，对肋骨骨折定位和诊断进行训练和验证。以高年资医师诊断为金标准，应用?字2分割检验和单因素方差分析比较低年资医师、DCNN模型和在DCNN模型辅助下的低年资医师诊断肋骨骨折的精确率、召回率、F1-score和诊断用时。统计DCNN模型诊断的假阳性和假阴性病例数量。结果：在300例测试集胸部CT图像中，共发现797处肋骨骨折，DCNN模型有22例假阳性病例和62例假阴性病例。低年资医师、DCNN模型和在DCNN模型辅助下的低年资医师诊断肋骨骨折的精确率（χ2=8.85，P=0.012）和召回率（χ2=43.2，P＜0.001）有明显差别。低年资医师诊断肋骨骨折的精确率（94.2%）低于DCNN模型（97.1%），在DCNN模型辅助下，低年资医师诊断的精确率有所增加（96.4%），DCNN模型和在DCNN模型辅助下低年资医师诊断的精确率无明显差别（96.4%）。低年资医师诊断肋骨骨折的召回率（84.8%）低于DCNN模型（92.2%），在DCNN模型辅助下医师诊断的召回率明显升高（94.0%）。低年资医师的诊断用时平均为（155.0±31.9）s，而DCNN模型诊断肋骨骨折仅需（4.8±1.4）s，在DCNN模型辅助下医师诊断用时可缩短至（40.6±7.0）s，三者有明显差别（F=328.1，P＜0.001）。结论：DCNN模型在胸部CT图像上可准确定位、诊断肋骨骨折，显著缩短诊断用时，减少漏诊、误诊率。

Abstract:

Objective： To explore the role of deep convolution neural network（DCNN） model in the automatic location and diagnosis rib fractures on chest CT images.　Methods：The images of 2 300 patients who underwent chest CT examinations because of initial thoracic trauma were enrolled retrospectively，300 cases were enrolled as test set.The DCNN model composed of segmentation network，key point detection network and fracture detection network，were used to train and validate the location and diagnosis of rib fractures. Taking the diagnosis of senior radiologists as the gold standard， ?字2 test and One-way ANOVA were used to compare the accuracy rate，recall rate，F1 score and the diagnosis time of junior radiologists，DCNN model and junior radiologists assisted by DCNN model in the diagnosis of rib fractures. The number of false positive and false negative cases diagnosed by DCNN model was counted. Results：　A total of 797 rib fractures were found in the chest CT images of the test set. There were 22 false positive cases and 62 false negative cases in DCNN model.The accuracy rate（χ2=8.85，P=0.012）and the recall rate（χ2=43.2，P＜0.001）among the junior radiologists，DCNN model and junior radiologists assisted by DCNN model had significant differences. The accuracy of rib fractures diagnosed by junior radiologists （94.2%） was lower than that of DCNN model （97.1%）. With the assistance of DCNN model，the diagnostic accuracy of junior radiologists increased （96.4%）. There was no significant difference between the accuracy of DCNN model and that of junior radiologists assisted by DCNN model（96.4%）. The recall rate of rib fractures diagnosed by junior radiologists （84.8%） was lower than that of DCNN model （92.2%）. The recall rate of rib fractures of junior radiologists assisted by DCNN model was increased dramatically （94.0%）. The average diagnosis time of junior radiologists was（155.0 ±31.9）s，while that of DCNN model was only（4.8 ±1.4）s. With the aid of DCNN model，the diagnosis time of doctors could be shortened to（40.6±7.0）s，there was significant difference among the three groups（F=328.1，P＜0.001）.　Conclusion：The DCNN model can accurately locate and diagnose rib fractures on chest CT images，significantly shorten the time of diagnosis，and reduce the rate of missed diagnosis and misdiagnosis.

参考文献/References:

[1] HAMILTON C，BARNETT L，TROP A，et al. Emergency department management of patients with rib fracture based on a clinical practice guideline[J]. Trauma Surg Acute Care Open，2017，2（1）：e000133.
[2] SHULZHENKO N O，ZENS T J，BEEMS M V，et al. Number of rib fractures thresholds independently predict worse outcomes in older patients with blunt trauma[J]. Surgery，2017，161（4）：1083-1089.
[3] SANO A. Rib radiography versus chest computed tomography in the diagnosis of rib fractures[J]. Thorac Cardiovasc Surg，2018，66（8）：693-696.
[4] LIN F C，LI R Y，TUNG Y W，et al. Morbidity，mortality，associated injuries，and management of traumatic rib fractures[J]. J Chin Med Assoc，2016，79（6）：329-334.
[5] 潘亚玲，王晗琦，陆勇. 人工智能在医学影像CAD中的应用[J]. 国际医学放射学杂志，2019，42（1）：3-7.
[6] ZHOU Q Q，WANG J S，TANG W，et al. Automatic detection and classification of rib fractures on thoracic CT using convolutional neural network：accuracy and feasibility[J]. Korean J Radiol，2020，21（7）：869-879.
[7] JIN L，YANG J，KUANG K，et al. Deep-learning-assisted detection and segmentation of rib fractures from CT scans： development and validation of FracNet[J]. EBioMedicine，2020，21（7）：869-879.
[8] HUANG H，YANG G，ZHANG W，et al.A deep multi-task learning framework for brain tumor segmentation[J]. Front Oncol，2021，11：690244.
[9] LU W，WEI J，XU T，et al.Quantitative CT for detecting COVID-19 pneumonia in suspected cases[J].BMC Infect Dis，2021，21（1）：836.
[10] GAN K，XU D，LIN Y，et al. Artificial intelligence detection of distal radius fractures： a comparison between the convolutional neural network and professional assessments[J]. Acta Orthop，2019，90（4）：394-400.
[11] KIM D H，MACKINNON T. Artificial intelligence in fracture detection：transfer learning from deep convolutional neural networks[J]. Clin Radiol，2018，73（5）：439-445.
[12] URAKAWA T，TANAKA Y，GOTO S，et al. Detecting intertrochanteric hip fractures with orthopedist-level accuracy using a deep convolutional neural network[J]. Skeletal Radiol，2019，48（2）：239-244.
[13] YU J S，YU S M，ERDAL B S，et al. Detection and localisation of hip fractures on anteroposterior radiographs with artificial intelligence： proof of concept[J]. Clin Radiol，2020，75（3）：231-237.
[14] KITAMURA G，CHUNG C Y，MOORE B E. Ankle fracture detection utilizing a convolutional neural network ensemble implemented with a small sample，de novo training，and multiview incorporation[J]. J Digit Imaging，2019，32（4）：672-677.
[15] CHUNG S W，HAN S S，LEE J W，et al. Automated detection and classification of the proximal humerus fracture by using deep learning algorithm[J]. Acta Orthop，2018，89（4）：468-473.
[16] PRANATA Y D，WANG K，WANG J，et al. Deep learning and SURF for automated classification and detection of calcaneus fractures in CT images[J]. Comput Methods Programs Biomed，2019，171：27-37.
[17] TOMITA N，CHEUNG Y Y，HASSANPOUR S. Deep neural networks for automatic detection of osteoporotic vertebral fractures on CT scans[J]. Comput Biol Med，2018，98： 8-15.

相似文献/References:

[1]骆源,徐启飞,吕泽政,等.三维ResNet 网络预测肺腺癌结节亚型的效能及其 稳定性[J].天津医科大学学报,2022,28(03):295.
　LUO Yuan,XU Qi-fei,LYU Ze-zheng,et al.Efficacy and stability of 3D ResNet for predicting nodule subtypes in lung adenocarcinoma[J].Journal of Tianjin Medical University,2022,28(02):295.

备注/Memo

备注/Memo:

基金项目 中国博士后科学基金面上二等资助（2019M651053）
作者简介 孟祥虹（1985-），女，副主任医师，博士，研究方向：骨与关节影像诊断；通信作者：马信龙，E-mail：maxinlong20190824@163.com。

常用功能

更新日期/Last Update: 2022-03-20