基于MPRAGE影像的影像组学列线图模型对颞叶癫痫与颞叶癫痫附加征鉴别的作用  

Differentiation of temporal lobe epilepsy and temporal plus epilepsy using radiomics nomogram based on MPRAGE images

作　　者：闫晓明[1] 阴方昭 遇涛[1] 张晓华[1] 张希 徐翠萍[1] 周晓霞[1] Yan Xiaoming;Yin Fangzhao;Yu Tao;Zhang Xiaohua;Zhang Xi;Xu Cuiping;Zhou Xiaoxia(Beijing Institute of Functional Neurosurgery,Xuanwu Hospital,Capital Medical University,Beijing 100053,China;Tianjin Huanhu Hospital,Tianjin 300350,China)

机构地区：[1]首都医科大学宣武医院功能神经外科,北京100053 [2]天津市环湖医院,天津300350

出　　处：《中华医学杂志》2024年第9期704-707,共4页National Medical Journal of China

基　　金：中国高校产学研创新基金(2021BCE01002);北京市自然科学基金(7232074)

摘　　要：本研究回顾性分析了2019年1月1日至2021年1月1日于北京宣武医院就诊的82例颞叶癫痫(TLE)和颞叶癫痫附加征(TPE)患者,男女各41例,年龄2~52(24±10)岁。通过Python随机分为训练集(58例)与测试集(24例)。采用FreeSurfer软件对患侧半球进行皮质分割,定义33个感兴趣区域(ROI),并通过Python提取影像组学特征。使用过滤式特征选择法筛选特征后,利用逻辑回归分类器构建影像组学模型并计算影像组学评分。结合临床特征与影像组学评分,通过R软件构建列线图模型,采用C指数评估模型预测准确性,并用Hosmer-Lemeshow方法检验模型拟合优度。结果显示,TLE与TPE患者在患病时间、颅内电极植入以及海马硬化方面差异有统计学意义(均P<0.05)。影像组学模型在训练集和测试集中的准确度分别为91.4%和87.5%。列线图模型应用C指数预测准确性。采用Hosmer-Lemeshow方法检验模型的拟合优度情况,在训练集和测试集中的AUC分别为0.95(95%CI:0.853~0.991)和0.84(95%CI:0.676~0.999)。本研究表明,基于三维磁化预处理快速采集梯度回波序列的影像组学列线图模型能有效鉴别TLE与TPE,为临床个体化治疗方案的制定提供参考。A total of 82 patients with temporal lobe epilepsy(TLE)and temporal plus epilepsy(TPE)admitted in Xuanwu Hospital from January 1,2019,to January 1,2021 were restrospectively analyzed,including 41 males and 41 females,aged 2 to 52(24±10)years.The patients were randomly divided into the training set(58 cases)and test set(24 cases)by Python.FreeSurfer software was used to segment the cortex of the affected hemisphere,defining 33 regions of interest(ROIs),and radiomics features were extracted by Python.After selecting features using the filter-based feature selection method,a radiomics model was constructed with a logistic regression classifier,and radiomics scores were calculated.Combining clinical characteristics with radiomics scores,a nomogram model was constructed using R software,the predictive accuracy of the model was assessed with the concordance index(C-index),and the model′s goodness-of-fit was tested with the Hosmer-Lemeshow method.The results showed statistically significant differences between TLE and TPE patients in disease duration,intracranial electrode implantation,and hippocampal sclerosis(both P<0.05).The accuracy of the radiomics model in the training set and the test set was 91.4%and 87.5%,respectively.The nomogram model uses C-index to predict accuracy.Hosmer-Lemeshow method was used to test the goodness of fit,with AUCs of 0.95(95%CI:0.853-0.991)in the training set and 0.84(95%CI:0.676-0.999)in the test set.The study indicates that the radiomics nomogram model based on MPRAGE sequences can effectively differentiate TLE from TPE,providing reference for the development of personalized treatment plans in clinical practice.

关 键 词：影像组学 机器学习 列线图 颞叶癫痫附加征 三维磁化预处理快速采集梯度回波序列 

分 类 号：R742.1[医药卫生—神经病学与精神病学]