基于Logistic回归建立磁共振成像临床效果评价评分模型的方法研究

Study on the method of establishing clinical evaluation model of MRI effect based on Logistic regression

目的磁共振成像(magnetic resonance imaging,MRI)临床应用广泛,但国内没有相应的临床效果评价标准,本研究通过Logistic回归模型建立一套标准化的临床效果评价标准,从而促进MR产业健康发展。材料与方法临床收集165张MR图像,评价每张图像中影响腰椎患者T2抑脂序列MRI质量的10个指标及MRI质量,通过Logistic回归科学建模,并采用H-L卡方检验和受试者工作特征曲线下面积(the area under the receiver-operating characteristic curve,AUC)检验模型的标定能力和区分能力。结果模型显示当总分低于3分时,MRI质量好的最高概率为0.02,认为MRI质量较差,不能应用于临床诊断,建议患者需重新拍摄MR图像;当总分在5~6分时,对应概率为0.22~0.52,认为MRI质量一般,勉强应用于临床诊断;当总分在8~9时,对应概率为0.94~0.98,认为MRI质量非常好,可很好地应用于临床诊断。H-L χ~2值为1.457(P=0.962),AUC为0.878(95%CI:0.814~0.941)。结论基于Logistic回归建立的MRI临床效果评价模型,具有很好的标定能力和区分能力,可很好地应用于临床。

Objective： MRI clinical application is extensive, but there is no corresponding clinical evaluation criteria. In this study, a set of standardized clinical effcacy evaluation criteria was established by Logistic regression model to promote the healthy development of MR industry. Materials and Methods： We collected 165 clinical MRI and ten factors infuencing the T2 lipid suppression sequence in lumbar vertebrae were collected of each image. The 10 variables of MRI mass were analyzed and the MRI quality was evaluated as the dependent variable. Logistic regression was used to scientifically model and evaluate the clinical effect of MRI. We used the H-L χ2 test and the AUC （the area under the receiver-operating characteristic curve） value to test the calibration and discrimination of the model. Results： The model shows that when the total score was less than 3 points, the highest probability of good MRI was 0.02, that MRI quality was poor, didn＇t be used in clinical diagnosis, it was recommended that patients need to re-shoot MRI. When the total score of 5-6 points, the corresponding probability was 0.22-0.52, that the general quality of MRI, barely applied to clinical diagnosis. When the total score of 8-9, that the MRI quality was very good, can be very accurate for clinical diagnosis. The H-L value was 1.457 （P=0.962）. The AUC value was 0.878 （95% CI： 0.814-0.941）. Conclusions： Based on Logistic regression, the evaluation model has good calibration ability and distinguishing ability, which can be used in clinical practice to establish a standardized standard of clinical evaluation of MRI.