基于影像组学的机器学习方法预测腰背筋膜炎

Prediction of Low Back Fasciitis by Machine Learning Method Based on Radiomics

摘要: 目的：腰背部筋膜炎是下背痛的常见但易被忽视的病因，早期识别筋膜的异常变化对于发现下背痛的隐匿病因至关重要。因此，本研究旨在采用基于腰椎软组织磁共振成像的影像组学机器学习方法，分析下背痛患者腰背部筋膜的变化特征。方法：我们对2021年9月至2023年9月期间380例腰痛患者的腰椎磁共振成像(MRI)进行了回顾性分析，收集临床资料。使用七种不同的分类算法建立基于Radscore的分类预测模型，并通过受试者工作特征(ROC)曲线评估其预测性能，选择表现最佳的模型作为Radiomics模型。将患者随机分配到训练组(70%, n = 266)或验证组(30%, n = 114)，分别建立临床模型、Radiomics模型和Nomogram模型，并通过ROC曲线评估其预测性能。结果：在七种机器学习模型中，Lasso模型的诊断性能最好，其曲线下面积(AUC)为0.83。因此，我们选择Lasso模型来构建影像组学模型。Nomogram模型结合了临床和影像组学特征。在训练集和验证集中，该模型表现良好，AUC分别为0.97和0.96。AUC和决策曲线分析(DCA)表明，Nomogram模型能够有效诊断腰筋膜炎。结论：综上所述，我们构建了一个基于临床特征和影像组学特征的Nomogram模型，旨在帮助临床医生通过软组织磁共振成像来识别和预测腰椎筋膜炎。

Abstract: Objective: Lumbar and dorsal fasciitis is a common but often overlooked cause of lower back pain. Early identification of abnormal changes in the fascia is crucial for discovering the hidden causes of lower back pain. Therefore, this study aims to use radiomics machine learning methods based on magnetic resonance imaging of lumbar soft tissues to analyze the changing characteristics of the lumbar and dorsal fascia in patients with lower back pain. Methods: We conducted a retrospective analysis of lumbar magnetic resonance imaging (MRI) of 380 patients with low back pain from September 2021 to September 2023 and collected clinical data. Seven different classification algorithms were used to establish a classification prediction model based on Radscore, and its prediction performance was evaluated through the receiver operating characteristic (ROC) curve. The model with the best performance was selected as the Radiomics model. Patients were randomly assigned to the training group (70%, n = 266) or the validation group (30%, n = 114). Clinical models, Radiomics models, and Nomogram models were established respectively, and their predictive performance was evaluated through the ROC curve. Result: Among the seven machine learning models, the Lasso model has the best diagnostic performance, with an area under the curve (AUC) of 0.83. Therefore, we chose the Lasso model to construct the radiomics model. The Nomogram model combines clinical and radiomics features. In the training set and validation set, the model performed well, with AUCs of 0.97 and 0.96 respectively. AUC and decision curve analysis (DCA) indicated that the Nomogram model could effectively diagnose lumbar fasciitis. Conclusion: In summary, we have constructed a Nomogram model based on clinical and radiomics features, aiming to assist clinicians in identifying and predicting lumbar fasciitis through soft tissue magnetic resonance imaging.