公共人工智能平台在膝关节骨性关节炎分期中的应用

Application of public artificial intelligence platform in staging of knee osteoarthritis

目的探讨利用X线片在公共人工智能平台上训练模型对膝关节骨性关节炎(KOA)严重程度自动分期的可行性。方法选取按照Kellgren-Lawrence(KL)分期系统进行分期的X线片,在公共人工智能平台上训练模型。最终使用了1445幅图像进行自动训练及测试评估。使用50幅图像的测试集对模型和放射科医师进行测试,计算放射科医师的准确率和F1-score,并与人工智能平台中模型返回的结果进行比较。结果模型对人工智能平台自动训练集的准确率为0.73,F1-score为0.72;模型对50幅图像的测试子集的准确率为0.70,F1-score为0.69。放射科医师测试的准确率为0.64,F1-score为0.63。模型效能达到甚至超过了高年资放射科医师测试水平。结论基于公共人工智能平台进行模型训练,利用X线图像进行KOA的自动KL分期,具有可行性和一定的优越性。

Objective To explore the feasibility of automatic grading of knee osteoarthritis(KOA)severity by using X-ray film training model on public artificial intelligence platform.Methods The Kellgren-Lawrence(KL)staging system was selected to determine stages of the X-ray films,and the model was trained on a public artificial intelligence platform.Finally,1445 images were used for automatic training and test evaluation.A test set of 50 images was used to test the model and the radiologists,and accuracy and F1-score of the radiologists were calculated and compared with the results returned by the model in the artificial intelligence platform.Results The accuracy of the model to the automatic training set of artificial intelligence platform was 0.73 and F1-score was 0.72;the accuracy of the model was 0.70 and F1-score was 0.69 for the test subset of 50 images;the accuracy of the radiologists test was 0.64 and F1-score was 0.63.Model performance matched or even exceeded that of senior radiologists.Conclusion It is feasible and advantageous to train the model based on public artificial intelligence platform and use X-ray image to perform automatic staging of KOA by KL.