AI-CAD辅助不同年资医生CT影像判读预测脑出血早期血肿扩大的效果分析  

Effect of AI-CAD assisting doctors with different seniority in CT image interpretation to predict the enlargement of hematoma in early stage of cerebral hemorrhage

作　　者：魏伟[1] 范文静 陈欣 张哲铭[1] 李国梁 陈东[1] WEI Wei;FAN Wenjing;CHEN Xin;ZHANG Zheming;LI Guoliang;CHEN Dong(Department of Neurosurgery,Central Hospital of Dalian University of Technology,Dalian,Liaoning 116033,China;Department of Epidemiology,Dalian Medical University,Dalian,Liaoning 116044,China)

机构地区：[1]大连理工大学附属中心医院神经外科,辽宁省大连市116033 [2]大连医科大学流行病学教研室,辽宁省大连市116044

出　　处：《中国动脉硬化杂志》2024年第5期429-436,共8页Chinese Journal of Arteriosclerosis

基　　金：大连市青年科技之星项目(2020RQ086);大连理工大学附属中心医院“登峰计划”院内自主立项项目(2022ZZ207)。

摘　　要：[目的]拟探讨人工智能(AI)辅助对不同年资医生预测脑出血早期血肿扩大的效果差异。[方法]回顾性地纳入大连理工大学附属中心医院诊断为脑出血的患者108例,收集入院时CT影像和入院后24 h CT影像,将病人入院时平扫CT获得的DICOM图像输入到Biomind与天坛合作开发的AI-CAD模型。在大连理工大学附属中心医院神经外科选择不同年资的医生共9名,先对患者进行独立预测,再结合辅助AI结果预测患者24 h内是否会出现血肿扩大。分别计算不同年资医生独立预测以及辅助AI预测脑出血早期血肿扩大的准确度,采用配对样本的McNemar检验不同医生间独立预测符合率和辅助AI预测准确度的差异显著性。[结果]高、中、低年资医生独立预测脑出血早期血肿扩大的准确度分别为58.95%、50.62%和38.89%,AI辅助后,预测准确度均显著提升(P<0.001),提升幅度最大的是低年资医生,为25.92%,其次是中年资医生,为19.75%,最小为高年资医生,为11.73%。在独立预测脑出血血肿扩大时,高年资医生灵敏度为18.75%(95%CI:9.44%~33.10%),特异度为65.94%(95%CI:59.98%~71.45%),中年资医生灵敏度为16.67%(95%CI:7.97%~30.76%),特异度为56.52%(95%CI:50.44%~62.42%),低年资医生灵敏度为8.33%(95%CI:2.70%~20.87%),特异度为44.20%(95%CI:38.29%~50.28%);但在AI辅助各年资医生预测后,各年资医生灵敏度和特异度均提高,高年资医生灵敏度为60.42%(95%CI:45.29%~73.88%),特异度为72.46%(95%CI:66.72%~77.57%),中年资医生灵敏度为64.58%(95%CI:49.40%~77.45%),特异度为71.38%(95%CI:65.59%~76.56%),低年资医生灵敏度为68.75%(95%CI:53.60%~80.91%),特异度为64.13%(95%CI:58.13%~69.73%)。[结论]AI-CAD辅助对高、中、低年资医生预测脑出血早期血肿扩大的准确性均有提升,尤其能显著提高低年资医生发现早期血肿扩大的能力,能够在一定程度上弥补低年资医生工作经验不足的问题。Aim To investigate the effect of artificial intelligence(AI)assisting doctors with different seniority in predicting the enlargement of hematoma in the early stage of cerebral hemorrhage.Methods A total of 108 patients diagnosed with cerebral hemorrhage in Central Hospital Affiliated to Dalian University of Technology were retrospectively collected.CT images at admission and 24 hours after admission were collected.DICOM images obtained from plain CT scan were input into AI-CAD model developed by Biomind in collaboration with Temple of Heaven.A total of 9 doctors of different senior-level were selected in neurosurgery department of our hospital.Firstly,independent prediction was applied in the patients and then the study predicted whether patients would delelop hematoma enlargement within 24 hours combined with the results of auxiliary AI.The accuracy of independent prediction of doctors with different seniority and assisted AI prediction of aneurysm stability was calculated respectively.McNemar of paired samples was used to test the significance of difference between independent prediction coincidence rate and assisted AI prediction accuracy among differ-ent doctors.Results The accuracy of high,middle and low seniority doctors independently predicting the early expan-sion of cerebral hemorrhage was 58.95%,50.62%and 38.89%,respectively,and the accuracy of prediction was signifi-cantly improved after assisted AI(P<0.001),the highest increase rate was low seniority doctors(25.92%),followed by middle seniority doctors(19.75%)and high seniority doctors(11.73%).The ability of senior physicians to independ-ently predict the expansion of intracerebral hemorrhage was strongest in patients and non-patients,with sensitivity of 18.75%(95%CI:9.44%~33.10%)and specificity of 65.94%(95%CI:59.98%~71.45%).The sensitivity of middle seniority doctors was 16.67%(95%CI:7.97%~30.76%),the specificity was 56.52%(95%CI:50.44%~62.42%),and the sensitivity of low seniority doctors was 8.33%(95%CI:2.70%~20.87%),the specificity was 44.20%(95%C

关 键 词：人工智能 脑出血 血肿扩大 

分 类 号：R743[医药卫生—神经病学与精神病学] R5[医药卫生—临床医学]