基于MRI影像组学在诊断前交叉韧带损伤分级中的应用

Application of MRI radiomics in classification diagnosis of anterior cruciate ligament injury

目的 分析MRI影像组学对前交叉韧带损伤分级的诊断价值。方法 回顾性选择2017~2022年滨州医学院附属医院放射科质子加权成像提示前交叉韧带异常信号患者212例,以关节镜检查结果为金标准,将患者分为重度损伤组(n=141)和轻度损伤组(n=71)。提取前交叉韧带图像的组学特征包括形状特征、一阶特征、纹理特征和小波特征。通过SMOTE法进行过采样,解决数据不均衡问题。通过组内及组间相关系数进行一致性检验。在两组中通过7:3的比例随机分割为训练集和测试集。并用LASSO算法筛选出最佳影像组学特征,采用Logistic回归建立组学模型。基于影像组学特征建立影像组学模型,基于两组的临床参数建立临床模型,以及联合两者建立Nomogram模型。分别在训练集及测试集绘制ROC曲线,计算敏感度、特异性、准确度评估模型的诊断效能,通过绘制校准曲线来评估模型预测值和实际观测值之间的差异,通过绘制临床决策曲线分析评价其临床有效性。结果 通过特征提取在质子加权成像横断位、冠状位、矢状位共获得2553个特征,通过特征筛选及降维最终保留12个特征参数。影像组学模型训练集的曲线下面积(AUC)为0.9105,测试组为0.8561。性别、关节不稳、关节交锁以及组学积分是诊断前交叉韧带重度损伤的最佳特征集。临床模型训练集的AUC值为0.6989,测试组为0.6415。Nomogram模型训练集的AUC值为0.9449,测试组为0.8661。Nomogram模型与临床模型差异有统计学意义(P<0.05),Nomogram的AUC高于组学模型,但在测试集差异无统计学意义(P>0.05)。结论 基于MRI的影像组学方法可为诊断前交叉韧带损伤分级提供一种新型的检测手段,使前交叉韧带损伤的临床诊断准确率得到很大提升,Nomogram模型比临床模型以及组学模型的诊断效能更好。

Objective To analyze the diagnostic value of MRI radiomics in the classification of anterior cruciate ligament(ACL)injury.Methods A total of 212 patients with abnormal signals of the ACL suggested by proton density weighted imaging in the Department of Radiology of the Affiliated Hospital of Binzhou Medical University from 2017 to 2022 were reviewed.The parents were divided into severe injury group(n=141)and mild injury group(n=71)based on arthroscopic findings,which was the gold standard in the classification of ACL injury.The radiomics features of the ACL images were extracted including shape features,first order features,texture features and wavelet features.The imbalance problem of data was solved by performing oversampling using SMOTE method.Consistency test was performed by intra-group and inter-group correlation coefficients.Patients were randomly divided into a training set and a test set at 7:3.The best radiomics features were selected by LASSO algorithm,and the radiomics model was established by Logistic regression.The radiomics model was established based on the radiomics features and the clinical model was established based on the clinical parameters of the two groups.The Nomogram model was established by the radiomics features and the clinical parameters as mentioned above.The diagnostic efficacy of the model was evaluated by ROC curves,sensitivity,specificity and accuracy in the training set and the test set.The difference between the model's predicted values and the actual observed values was evaluated by the calibration curve,and the clinical efficacy was evaluated by the clinical decision curve analysis.Results A total of 2553 features were obtained from proton density weighted imaging transverse,coronal and sagittal positions through feature extraction,and 12 feature parameters were retained through feature filtering and dimensionality reduction.The area under the curve(AUC)was 0.9105 for the training set of the radiomics model and 0.8561 for the test set.The best feature sets of diagnosing severe ACL injury were gender,joint instability,joint interlock and radiomics.In the clinical model,the AUC values of the training set was 0.6989 and the test set was 0.6415.In the Nonogram model,the AUC values of the training set was 0.9449 and the test set was 0.8661.The difference between the Nomogram model and the clinical model was statistically significant(P<0.05).The AUC of the Nomogram was higher than that of the radiomics model,but the difference was not statistically significant in the test set(P>0.05).Conclusion The radiomics method based on MRI images can provide a new diagnostic method in classification of ACL injury which can greatly improve the clinical diagnosis accuracy of ACL injury.Furthermore,the diagnostic efficiency of the Nomogram model is the best compared with the clinical model and the radiomics model.