基于深度学习成人胸部X线摄影自动分类的临床应用价值

The clinical application value of automatic classification based on the deep learning of adult chest X-ray photography

目的探讨深度学习胸部X线摄影片辅助诊断系统对成人X线摄影胸片征象的诊断效能及自动分类中的应用价值。方法选取我院就诊并且行胸部X线摄影平片检查的763例患者。采用Dr.Wise胸部X线摄影片辅助诊断系统检测疾病征象并自动分类胸片优先级,包括正常、可疑异常和危急分类。将诊断报告结果作为标准,测试人工智能辅助诊断系统检测正常胸片、一般异常、危急征象的诊断效能,包括敏感度、特异度、准确率。将诊断报告结果与辅助诊断系统结果相比较,并采用Kappa检验计算两者的诊断一致性。计算使用AI辅助诊断系统后,危急征象胸片的诊断时间。结果763例患者中按照诊断标准,三个分类分别为:正常280例,一般异常374例,危急征象109例。统计正常胸片的诊断效能:敏感度94.6%,特异度95.7%,准确率95.3%;一般异常胸片的诊断效能:敏感度为84.8%,特异度为83.0%,准确率为84.0%;危急胸片的诊断效能:敏感度为91.7%,特异度为96.8%,准确率96.1%。辅助诊断系统与影像诊断对于正常、危急的诊断一致性较高,kappa值约0.749、0.735(P<0.01)。原报告平均时间约为(55.6±23.7)min。使用AI辅助诊断系统后,危急报告的诊断报告时间约为(10.3±5.6)min。结论深度学习胸部X线摄影片辅助诊断系统对危急征象的诊断敏感度较高,能有效对胸片进行诊断分类,缩短危急报告诊断时间,优化诊断流程。

Objective To explore the effectiveness of deep learning chest radiography auxiliary diagnosis system to the diagnosis of adult X-ray photographysigns and the application value of the automatic classification.Methods 763 patients who underwent plain chest X-ray photography examination were selected and Dr.Wise’chest X-ray film auxiliary diagnosis system was adopted to automatically detect signs of disease and to classify the sternum priorities,including normal and abnormal suspicious and emergency.The diagnostic results were compared with those by artificial intelligent auxiliary diagnosis system and the consistency between the two diagnostic modalities was calculated using Kappa test.The diagnostic time of emergency X-ray signs was calculated after using the AI auxiliary diagnostic system.Results According ccording to diagnostic criteria,763 patients were divided into three classifications respectively,i.e.,there were 280 cases of normality,374 of general abnormality,and 109 of appearing emergency signs.For the effectiveness of the diagnosis of normal chest radiograph,the sensitivity was 94.6%,the specificity was 95.7%,and the accuracy was 95.3%;For the efficiency of average abnormal chest X-ray diagnosis,the sensitivity was 84.8%,specificity was 83.0%,and accuracy was 84.0%.As for the effectiveness of diagnosis of critical radiograph,the sensitivity was 91.7%,specificity was 96.8%,and accuracy was 96.1%.Auxiliary diagnostic system and the imaging diagnosis exhibited high consistency for normal and emergency diagnosis.Kappa value was about 0.749,0.735(P<0.01).The average time of original report was(55.6±23.7)min.After using AI auxiliary diagnosis system,critical report diagnostic time was about(10.3±5.6)min.Conclusion The sensitivity of deep learning chest radiography auxiliary diagnosis system is higher,which can effectively classify chest X-ray diagnosis,shorten the time of critical report diagnosis,and optimize the diagnostic workflow.