切换至 "中华医学电子期刊资源库"

中华医学超声杂志(电子版) ›› 2022, Vol. 19 ›› Issue (03) : 256 -261. doi: 10.3877/cma.j.issn.1672-6448.2022.03.012

基础研究

人工智能辅助床旁超声诊断肾创伤的基础研究
马骏1, 罗渝昆2,(), 何雪磊3, 高菡静2, 王坤4, 宋青2, 王妍洁2, 陈淑媛2, 余桂花2   
  1. 1. 100853 北京,解放军医学院;100853 北京,解放军总医院第一医学中心超声科
    2. 100853 北京,解放军总医院第一医学中心超声科
    3. 710127 西安,西北大学信息科学与技术学院
    4. 100190 北京,中科院自动化研究所分子影像重点实验室
  • 收稿日期:2021-12-13 出版日期:2022-03-01
  • 通信作者: 罗渝昆

Artificial intelligence aided point of care ultrasound for diagnosis of renal injury: a pilot study

Jun Ma1, Yukun Luo2,(), Xuelei He3, Hanjing Gao2, Kun Wang4, Qing Song2, YanJie Wang2, Shuyuan Chen2, Guihua Yu2   

  1. 1. Department of Ultrasound, Medical School of Chinese PLA, Beijing 100853, China; Department of Ultrasound, First Medical Center, Chinese PLA General Hospital, Beijing 100853, China
    2. Department of Ultrasound, First Medical Center, Chinese PLA General Hospital, Beijing 100853, China
    3. School of Information Sciences and Technology, Northwest University, Xi'an 710127, China
    4. CAS Key Laboratory of Molecular Imaging, Institute of Automation, Chinese Academy of Sciences, Beijing 100190, China
  • Received:2021-12-13 Published:2022-03-01
  • Corresponding author: Yukun Luo
引用本文:

马骏, 罗渝昆, 何雪磊, 高菡静, 王坤, 宋青, 王妍洁, 陈淑媛, 余桂花. 人工智能辅助床旁超声诊断肾创伤的基础研究[J/OL]. 中华医学超声杂志(电子版), 2022, 19(03): 256-261.

Jun Ma, Yukun Luo, Xuelei He, Hanjing Gao, Kun Wang, Qing Song, YanJie Wang, Shuyuan Chen, Guihua Yu. Artificial intelligence aided point of care ultrasound for diagnosis of renal injury: a pilot study[J/OL]. Chinese Journal of Medical Ultrasound (Electronic Edition), 2022, 19(03): 256-261.

目的

探讨基于卷积神经网络(CNN)构建的人工智能辅助诊断模型对肾钝性创伤超声诊断的应用价值。

方法

建立不同程度动物肾创伤模型,通过床旁超声仪采集正常肾及创伤肾超声图片,分成训练集及测试集,根据造模位置和超声造影结果,手动勾画出肾轮廓,采用3折交叉验证进行分类训练及测试。绘制受试者工作特征(ROC)曲线,计算人工智能辅助诊断模型的敏感度、特异度、准确性和曲线下面积(AUC)。

结果

采集正常肾图片共1737张,各级别创伤肾图片共2125张,经过对测试集的验证,该模型可自动对肾创伤有无进行分类,对肾创伤诊断的平均敏感度为73%、平均特异度为85%、平均准确性为79%、AUC为0.80,诊断价值较高。

结论

基于CNN构建的深度学习模型辅助床旁超声仪在诊断肾创伤有无分类中取得了较满意的结果。

Objective

To explore the application value of artificial intelligence aided diagnosis model based on convolutional neural network (CNN) in ultrasonic diagnosis of blunt renal trauma.

Methods

Rabbits were used to simulate different grades of renal trauma model of renal trauma of different degrees was established. The ultrasonic images of the normal kidney and renal trauma were collected by point of care ultrasound (POCUS) and divided into either a training or a test cohort. According to the modeling position and contrast-enhanced ultrasound results, the renal contour was manually drawn and classified for training, followed by 3-fold cross-validation testing. The sensitivity, specificity, accuracy, and area under curve (AUC) of the artificial intelligence aided diagnosis model were calculated.

Results

A total of 1737 images of the normal kidney and 2125 images of traumatic kidney were collected. After the verification of the test set, the model can automatically classify the presence or absence of renal trauma. The average sensitivity for renal trauma diagnosis was 73%, the average specificity was 85%, the average accuracy was 79%, and the AUC was 0.80.

Conclusion

The deep learning assisted POCUS model constructed based on CNN has achieved satisfactory results in the diagnosis and classification of renal trauma.

表1 肾创伤建模程度与美国创伤协会肾创伤分级对比
图1 美国创伤协会肾创伤分级为Ⅰ级的超声图像。图a为常规二维图像,图b为超声造影图像,箭头所指处为包膜下血肿
图2 美国创伤协会肾创伤分级为Ⅱ级超声图像。图a为常规二维图像,图b为超声造影图像,箭头所指处为肾创伤灶,局限于肾皮质内,未伤及集合系统
图3 美国创伤协会肾创伤分级为Ⅲ~Ⅳ级超声图像。图a为常规二维图像,图b为超声造影图像;箭头所指处为创伤灶,创伤贯穿皮髓质,累及集合系统
图4 美国创伤协会肾创伤分级为5级超声图像。图a为常规二维图像,图b为超声造影图像;箭头所指处为创伤灶,肾碎裂,解剖结构难以辨认
图5 通过图像标签工具标记软件勾勒肾轮廓
图6 基于NASNet构建的模型结构图。卷积主干1和2里包括左右两个分支,每个分支有4层卷积层,在主干模块中进行特征拼接,然后输出到后两个卷积分支(0~11)中进行特征提取,最后将图片特征信息传输到12层用于训练参数进行平均池化拉伸与全链接输出
图7 模型整体框架示意图。将训练集超声图片进行3折训练后,调整参数得出动物肾创伤诊断模型
表2 3折交叉验证训练集所得诊断效能指标及其平均值
表3 3折交叉验证测试集所得诊断效能指标及其平均值
图8 训练集(图a)及测试集(图b)诊断肾损伤的受试者操作特征(ROC)曲线注:AUC为曲线下面积
1
James SL, Castle CD, Dingels ZV, et al. Global injury morbidity and mortality from 1990 to 2017: results from the Global Burden of Disease Study 2017 [J]. Inj Prev, 2020, 26(Supp 1): i96-i114.
2
Terrier JE, Paparel P, Gadegbeku B, et al. Genitourinary injuries after traffic accidents: Analysis of a registry of 162,690 victims [J]. J Trauma Acute Care Surg, 2017, 82(6): 1087-1093.
3
Erlich T, Kitrey ND. Renal trauma: the current best practice [J]. Ther Adv Urol, 2018, 10(10): 295-303.
4
Lin Q, Lv F, Luo Y, et al. Contrast-enhanced ultrasound for evaluation of renal trauma during acute hemorrhagic shock: a canine model [J]. J Med Ultrason (2001), 2015, 42(2): 199-205.
5
Cao ZQ, Liang M, Li SX, et al. A new method for traumatic renal injury in a canine model [J]. J Chin Med Assoc, 2017, 80(3): 133-139.
6
Khoschnau S, Jabbour G, Al-Hassani A, et al. Traumatic kidney injury: an observational descriptive study [J]. Urol Int, 2020, 104(1-2): 148-155.
7
Holmes JF, Harris D, Battistella FD. Performance of abdominal ultrasonography in blunt trauma patients with out-of-hospital or emergency department hypotension [J]. Ann Emerg Med, 2004, 43(3): 354-361.
8
全军超声医学专业委员会, 中国医学装备协会应急救治装备分会. 战术战伤救治中的超声技术应用专家共识 [J/CD]. 中华医学超声杂志(电子版), 2019, 12(16): 892-898.
9
Stengel D, Leisterer J, Ferrada P, et al. Point-of-care ultrasonography for diagnosing thoracoabdominal injuries in patients with blunt trauma [J]. Cochrane Database Syst Rev, 2018, 12(12): Cd012669.
10
Waheed KB, Baig AA, Raza A, et al. Diagnostic accuracy of focused assessment with sonography for trauma for blunt abdominal trauma in the Eastern Region of Saudi Arabia [J]. Saudi Med J, 2018, 39(6): 598-602.
11
Valentino M, Serra C, Zironi G, et al. Blunt abdominal trauma: emergency contrast-enhanced sonography for detection of solid organ injuries [J]. AJR Am J Roentgenol, 2006, 186(5): 1361-1367.
12
李志艳, 唐杰, 罗渝昆, 等. 超声造影评价肾创伤程度的相关因素分析 [J]. 中华超声影像学杂志, 2012, 21(4): 348-351.
13
李志艳, 唐杰, 罗渝昆, 等. 超声造影定量观察实验兔肾脏外伤不同程度活动性出血的初步研究 [J/CD]. 中华医学超声杂志(电子版), 2010, 7(10): 1646-1653.
14
李志艳, 唐杰, 吕发勤, 等. 超声造影引导经皮注射止血剂治疗肾挫裂伤出血 [J]. 中华超声影像学杂志, 2009, 18(12): 1053-1056.
15
Ramsingh D, Ma M, Le DQ, et al. Feasibility evaluation of commercially available video conferencing devices to technically direct untrained nonmedical personnel to perform a rapid trauma ultrasound examination [J]. Diagnostics (Basel), 2019, 9(4): 188.
16
Wydo SM, Seamon MJ, Melanson SW, et al. Portable ultrasound in disaster triage: a focused review [J]. Eur J Trauma Emerg Surg, 2016, 42(2): 151-159.
17
Sarkisian AE, Khondkarian RA, Amirbekian NM, et al. Sonographic screening of mass casualties for abdominal and renal injuries following the 1988 Armenian earthquake [J]. J Trauma, 1991, 31(2): 247-250.
18
Zhou J, Huang J, Wu H, et al. Screening ultrasonography of 2,204 patients with blunt abdominal trauma in the Wenchuan earthquake [J]. J Trauma Acute Care Surg, 2012, 73(4): 890-894.
19
Shorter M, Macias DJ. Portable handheld ultrasound in austere environments: use in the Haiti disaster [J]. Prehosp Disaster Med, 2012, 27(2): 172-177.
20
Brooks AJ, Price V, Simms M, et al. Handheld ultrasound diagnosis of extremity fractures [J]. J R Army Med Corps, 2004, 150(2): 78-80.
21
Tai N, Hill P, Kay A, et al. Forward trauma surgery in Afghanistan: lessons learnt on the modern asymmetric battlefield [J]. J R Army Med Corps, 2008, 154(1): 14-18.
22
Lecun Y, Bengio Y, Hinton G. Deep learning [J]. Nature, 2015, 521(7553): 436-444.
23
Erickson BJ, Korfiatis P, Akkus Z, et al. Machine learning for medical imaging [J]. Radiographics, 2017, 37(2): 505-515.
24
Jafari MH, Girgis H, Van Woudenberg N, et al. Automatic biplane left ventricular ejection fraction estimation with mobile point-of-care ultrasound using multi-task learning and adversarial training [J]. Int J Comput Assist Radiol Surg, 2019, 14(6): 1027-1037.
25
Jin Z, Zhu Y, Zhang S, et al. Ultrasound computer-aided diagnosis (CAD) based on the thyroid imaging reporting and data system (TI-RADS) to distinguish benign from malignant thyroid nodules and the diagnostic performance of radiologists with different diagnostic experience [J]. Med Sci Monit, 2020, 26: e918452.
26
Poletti PA, platon A, Becker CD, et al. Blunt abdominal trauma: does the use of a second-generation sonographic contrast agent help to detect solid organ injuries? [J]. AJR Am J Roentgenol, 2004, 183(5): 1293-1301.
27
Poletti PA, Kinkel K, Vermeulen B, et al. Blunt abdominal trauma: should US be used to detect both free fluid and organ injuries? [J]. Radiology, 2003, 227(1): 95-103.
28
Yi J, Kang HK, Kwon JH, et al. Technology trends and applications of deep learning in ultrasonography: image quality enhancement, diagnostic support, and improving workflow efficiency [J]. Ultrasonography, 2021, 40(1): 7-22.
29
王勇, 何文. 中国超声医学的发展与展望 [J]. 中国医学影像学杂志, 2019, 7(27): 479-480.
[1] 李洋, 蔡金玉, 党晓智, 常婉英, 巨艳, 高毅, 宋宏萍. 基于深度学习的乳腺超声应变弹性图像生成模型的应用研究[J/OL]. 中华医学超声杂志(电子版), 2024, 21(06): 563-570.
[2] 明昊, 肖迎聪, 巨艳, 宋宏萍. 乳腺癌风险预测模型的研究现状[J/OL]. 中华乳腺病杂志(电子版), 2024, 18(05): 287-291.
[3] 沈岩. 创伤腹膜后血肿的诊治策略[J/OL]. 中华危重症医学杂志(电子版), 2024, 17(06): 1-1.
[4] 叶莉, 杜宇. 深度学习在牙髓根尖周病临床诊疗中的应用[J/OL]. 中华口腔医学研究杂志(电子版), 2024, 18(06): 351-356.
[5] 熊鹰, 林敬莱, 白奇, 郭剑明, 王烁. 肾癌自动化病理诊断:AI离临床还有多远?[J/OL]. 中华腔镜泌尿外科杂志(电子版), 2024, 18(06): 535-540.
[6] 李伟, 宋子健, 赖衍成, 周睿, 吴涵, 邓龙昕, 陈锐. 人工智能应用于前列腺癌患者预后预测的研究现状及展望[J/OL]. 中华腔镜泌尿外科杂志(电子版), 2024, 18(06): 541-546.
[7] 黄俊龙, 李文双, 李晓阳, 刘柏隆, 陈逸龙, 丘惠平, 周祥福. 基于盆底彩超的人工智能模型在女性压力性尿失禁分度诊断中的应用[J/OL]. 中华腔镜泌尿外科杂志(电子版), 2024, 18(06): 597-605.
[8] 莫淇舟, 苏劲, 黄健, 李健维, 李思宁, 柳建军. 智能控压输尿管软镜碎石吸引取石术在直径10~25 mm上尿路结石中的应用[J/OL]. 中华腔镜泌尿外科杂志(电子版), 2024, 18(05): 497-502.
[9] 李义亮, 苏拉依曼·牙库甫, 麦麦提艾力·麦麦提明, 克力木·阿不都热依木. 机器人与腹腔镜食管裂孔疝修补术联合Nissen 胃底折叠术短期疗效分析[J/OL]. 中华疝和腹壁外科杂志(电子版), 2024, 18(05): 512-517.
[10] 赵毅, 李昶田, 唐文博, 白雪婷, 刘荣. 腹腔镜术中超声主胰管自动识别模型的临床应用[J/OL]. 中华腔镜外科杂志(电子版), 2024, 17(05): 290-294.
[11] 王守森, 傅世龙, 鲜亮, 林珑. 深入理解控制性减压技术对创伤性颅脑损伤术中脑膨出的预防机制与效果[J/OL]. 中华神经创伤外科电子杂志, 2024, 10(05): 257-262.
[12] 王如海, 王绅, 张敏, 李春, 韩超, 于强, 胡海成, 李习珍. 重型创伤性脑损伤患者去骨瓣减压术后短期死亡风险的影响因素分析[J/OL]. 中华脑科疾病与康复杂志(电子版), 2024, 14(05): 285-291.
[13] 司楠, 孙洪涛. 创伤性脑损伤后肾功能障碍危险因素的研究进展[J/OL]. 中华脑科疾病与康复杂志(电子版), 2024, 14(05): 300-305.
[14] 孙铭远, 褚恒, 徐海滨, 张哲. 人工智能应用于多发性肺结节诊断的研究进展[J/OL]. 中华临床医师杂志(电子版), 2024, 18(08): 785-790.
[15] 董晟, 郎胜坤, 葛新, 孙少君, 薛明宇. 反向休克指数乘以格拉斯哥昏迷评分对老年严重创伤患者发生急性创伤性凝血功能障碍的预测价值[J/OL]. 中华临床医师杂志(电子版), 2024, 18(06): 541-547.
阅读次数
全文


摘要