Hu Hai, Ye Long, Wu Ping, Shi Zhuowei, Chen Guangwen, Li Yongmei
Department of Radiology, The First Affiliated Hospital of Chongqing Medical University, Chongqing, China.
Department of Radiology, Chengdu Second People's Hospital, Chengdu, China.
Eur Radiol. 2025 Jun 17. doi: 10.1007/s00330-025-11771-2.
The study aimed to identify factors influencing the evolution of chronic lesions in multiple sclerosis (MS) using a machine learning approach.
Longitudinal data were collected from individuals with relapsing-remitting multiple sclerosis (RRMS). The "iron rim" sign was identified using quantitative susceptibility mapping (QSM), and microstructural damage was quantified via T1/fluid attenuated inversion recovery (FLAIR) ratios. Additional data included baseline lesion volume, cerebral T2-hyperintense lesion volume, iron rim lesion volume, the proportion of iron rim lesion volume, gender, age, disease duration (DD), disability and cognitive scores, use of disease-modifying therapy, and follow-up intervals. These features were integrated into machine learning models (logistic regression (LR), random forest (RF), and support vector machine (SVM)) to predict lesion volume change, with the most predictive model selected for feature importance analysis.
The study included 47 RRMS individuals (mean age, 30.6 ± 8.0 years [standard deviation], 6 males) and 833 chronic lesions. Machine learning model development results showed that the SVM model demonstrated superior predictive efficiency, with an AUC of 0.90 in the training set and 0.81 in the testing set. Feature importance analysis identified the top three features were the "iron rim" sign of lesions, DD, and the T1/FLAIR ratios of the lesions.
This study developed a machine learning model to predict the volume outcome of MS lesions. Feature importance analysis identified chronic inflammation around the lesion, DD, and the microstructural damage as key factors influencing volume change in chronic MS lesions.
Question The evolution of different chronic lesions in MS exhibits variability, and the driving factors influencing these outcomes remain to be further investigated. Findings A SVM learning model was developed to predict chronic MS lesion volume changes, integrating lesion characteristics, lesion burden, and clinical data. Clinical relevance Chronic inflammation surrounding lesions, DD, and microstructural damage are key factors influencing the evolution of chronic MS lesions.
本研究旨在采用机器学习方法确定影响多发性硬化症（MS）慢性病变演变的因素。
收集复发缓解型多发性硬化症（RRMS）患者的纵向数据。使用定量磁化率成像（QSM）识别“铁环”征，并通过T1/液体衰减反转恢复（FLAIR）比值对微观结构损伤进行量化。其他数据包括基线病变体积、脑T2高信号病变体积、铁环病变体积、铁环病变体积比例、性别、年龄、病程（DD）、残疾和认知评分、疾病修饰治疗的使用情况以及随访间隔。将这些特征整合到机器学习模型（逻辑回归（LR）、随机森林（RF）和支持向量机（SVM））中以预测病变体积变化，并选择预测性最强的模型进行特征重要性分析。
该研究纳入了47例RRMS患者（平均年龄30.6±8.0岁[标准差]，6例男性）和833个慢性病变。机器学习模型开发结果表明，SVM模型具有更高的预测效率，训练集的AUC为0.90，测试集的AUC为0.81。特征重要性分析确定前三个特征为病变的“铁环”征、病程和病变的T1/FLAIR比值。
本研究开发了一种机器学习模型来预测MS病变的体积转归。特征重要性分析确定病变周围的慢性炎症、病程和微观结构损伤是影响慢性MS病变体积变化的关键因素。
问题MS中不同慢性病变的演变具有变异性，影响这些转归的驱动因素仍有待进一步研究。发现开发了一种SVM学习模型来预测慢性MS病变体积变化，整合了病变特征、病变负荷和临床数据。临床意义病变周围的慢性炎症、病程和微观结构损伤是影响慢性MS病变演变的关键因素。
