基于深度学习的X线腰椎正侧位片的智能质量控制研究

Research on intelligent quality control of lumbar spine radiograph based on deep learning

目的:探讨基于深度学习的X线腰椎正侧位片的智能质量控制方法在临床工作中应用的可行性。方法:回顾性分析4690例X线腰椎正侧位片。将图像质量特征分为投照技术质量和图像清晰度质量2种类别,构建质控知识图谱。通过数据标注和模型训练,使用平均精确度均值(mAP)和平均绝对误差(MAE)评估智能质控与人工质控之间的差异。结果:根据参考标准,在投照技术质量方面,智能质控的mAP显著优于单人质控(A、B),与多人质控结果接近;在图像清晰度方面,智能质控的MAE显著小于单人及多人质控。结论:基于深度学习的X线腰椎正侧位片的智能质量控制方法优于单人质控,在图像清晰度质量方面也优于多人质控。该方法可客观评估图像质量,提高工作效率,有望在临床质控中推广应用。

Objective:To explore the feasibility of intelligent quality control method of lumbar spine X-rays based on deep learning technology in clinical work.Methods:A retrospective analysis of 4,690 cases of X-ray lumbar spine anteroposterior and lateral images was conducted.The image quality characteristics were classified into two categories,the image technical quality and image clarity quality,and a quality control knowledge graph was constructed.Through data annotation and model training,the differences between intelligent quality control and manual quality control were evaluated using mean average precision(mAP)and mean absolute error(MAE).Results:According to the reference standard,in terms of the image technical quality,the mAP of intelligent quality control method was significantly better than that of single-person quality control(A,B)and closed to that of multi-person quality control method.In terms of image clarity,the MAE of intelligent quality control method was significantly better than that of single-person and multi-person quality control methods.Conclusions:The intelligent quality control method for lumbar spine X-rays based on deep learning is superior to single-person quality control method,and also superior in image clarity quality to multi-person quality control method.This method can objectively evaluate image quality,improve work efficiency,and will be the potential for widespread application in clinical quality control.