基于AI与人工阅片对肋骨骨折性质诊断价值的比较

Comparison of the Diagnostic Value of Rib Fracture Properties Based on AI and Manual Reading

目的探究基于深度学习的人工智能(Artificial Intelligence,AI)技术在不同分类的肋骨骨折中的诊断价值。方法回顾性分析77例因胸部外伤导致的肋骨骨折患者的CT图像,首先由两名低年资影像医师及uAI EasyTriage-Rib软件(上海联影医疗公司)分别单独阅片,之后医师使用uAI EasyTriage-Rib软件再次阅片,记录每次阅片时患者的肋骨骨折点、骨折类型及诊断用时。对不同类型肋骨骨折(错位性骨折、非错位性骨折、陈旧性骨折)的检出率、检出准确度、灵敏度及AI诊断中出现的假阳性、假阴性情况进行统计学分析。结果uAI EasyTriage-Rib软件对错位性骨折、非错位性骨折、陈旧性骨折的检出率分别为96.04%、97.02%、98.79%。uAI EasyTriage-Rib软件对3种类型的肋骨骨折的诊断准确度、灵敏度均高于低年资影像科医师(P<0.05);影像科医师借助AI后对不同类型的肋骨骨折诊断准确度高于AI单独诊断(P<0.05);使用uAI EasyTriage-Rib软件的诊断用时显著低于医师单独阅片(P<0.05);uAI EasyTriage-Rib软件对不同类型的肋骨骨折总漏诊率为2.53%。结论AI可以有效地提升影像科医师对不同类型肋骨骨折的诊断效能,具有较高的临床应用价值,且医师与AI联合诊断效能更优。

Objective To study the diagnostic value of artificial intelligence(AI)technology based on deep learning in different classification of rib fractures.Methods The CT images of 77 patients with rib fracture caused by chest trauma were analyzed retrospectively.First,two junior radiologists and uAI EasyTriage-Rib software(Shanghai United Imaging Medical Co.,Ltd.)read the images separately,and then the radiologists used uAI EasyTriage-Rib software read the images again.The fracture point and fracture type were recorded,together with the diagnosis time.The detection rate,detection accuracy and sensitivity of different types of rib fractures and the false positive and false negative in AI diagnosis were analyzed.Results The detection rate of uAI EasyTriage-Rib software for different types of rib fractures were 96.04%(dislocation fracture),97.02%(non-dislocation fracture),98.79%(old fracture).The diagnostic accuracy and sensitivity of uAI EasyTriage-Rib software for different types of rib fractures were higher than those of low-seniority radiologists(P<0.05).The diagnostic accuracy of radiologists using AI for different types of rib fractures was higher than that of AI alone(P<0.05),and the reading time of using uAI EasyTriage-Rib software was significantly lower than that of radiologists alone(P<0.05).The false rate in the diagnosis of all types of rib fractures with uAI EasyTriage-Rib software were 2.53%.Conclusion AI can improve the diagnostic efficiency of radiologists effectively for different types of rib fractures and has great clinical application value.Joint diagnosis by physician and AI is more effective.