计算机辅助诊断在尘肺病诊断中应用价值

Application value of computer-aided diagnosis in diagnosing pneumoconiosis

目的探讨基于深度残差网络的计算机辅助诊断技术在职业性尘肺病(以下简称"尘肺病")诊断中的应用价值。方法采用方便抽样方法,收集5424例职业健康检查者数字化X射线胸片图像建立数据集。利用数据集对尘肺病计算机辅助诊断系统进行训练后,对测试集(尘肺病阳性、阴性病例各50例)进行独立诊断并输出阳性概率值。由6名不同年资诊断医师对测试集图像分别进行独立诊断和参考计算机结果进行辅助诊断。采用受试者工作特征曲线下面积(AUC值)、灵敏度和特异度对诊断准确性进行评价;采用Kappa一致性检验对诊断一致性进行评价。结果与使用计算机辅助诊断前比较,使用计算机辅助诊断后尘肺病诊断医师的AUC值、灵敏度、特异度、Kappa值均有所升高;其中,灵敏度从0.74提高到0.85(P<0.05),Kappa平均值从0.64提高到0.79(P<0.05);AUC值从0.90提高到0.95,特异度从0.89提高到0.94,但差异均无统计学意义(P>0.05)。结论计算机辅助诊断可提高尘肺病诊断医师尘肺病筛查的灵敏度和一致性,减少医师之间的诊断差异。

Objective To explore the application value of computer-aided diagnosis technology based on deep residual network in the diagnosis of occupational pneumoconiosis(hereinafter referred to as pneumoconiosis).Methods A total of 5424 digital radiography chest images were collected from occupational health examiners using a convenient sampling method.These images were used to establish a data set.After training with the data set,the pneumoconiosis computer-aided diagnosis system was used to independently diagnose the test set images(50 positive and negative cases each)and output a positive probability value.Six diagnostic physicians with varied ages and different experiences performed independent diagnosis on the test set and assisted diagnosis with reference to computer results.The diagnostic accuracy was evaluated using the area under the receiver operating characteristic curve(AUC)value,sensitivity,and specificity.The Kappa consistency test was used to evaluate the diagnostic consistency.Results The AUC value,sensitivity,specificity,and Kappa value of pneumoconiosis diagnosis increased after using computer-aided diagnosis.The sensitivity increased from 0.74 to 0.85(P<0.05)and the Kappa value increased from 0.64 to 0.79(P<0.05).The AUC value increased from 0.90 to 0.95,and the specificity increased from 0.89 to 0.94,but there were no statistical difference(P<0.05).Conclusion Computer-aided diagnosis can improve the sensitivity and consistency of pneumoconiosis screening and reduce the differences in diagnosis among physicians.