Cai Bingyang, Jiang Shize, Huang Hui, Li Jiwei, Yuan Siyu, Cui Ya, Bao Weiqi, Hu Jie, Luo Jie, Chen Liang
From the National Engineering Research Center of Advanced Magnetic Resonance Technologies for Diagnosis and Therapy, School of Biomedical Engineering, (B.C., H.H., J.L., S.Y., Y.C., J.L.), Shanghai Jiao Tong University, Shanghai, China.
Department of Neurosurgery (S.J., J.H., L.C.), Huashan Hospital, Fudan University, Shanghai, China.
AJNR Am J Neuroradiol. 2025 Jul 1;46(7):1493-1500. doi: 10.3174/ajnr.A8647.
Epilepsy, a globally prevalent neurologic disorder, necessitates precise identification of the epileptogenic zone (EZ) for effective surgical management. While the individual utilities of FDG-PET and flumazenil (FMZ)-PET have been demonstrated, their combined efficacy in localizing the epileptogenic zone remains underexplored. We aim to improve the noninvasive prediction of EZ in temporal lobe epilepsy (TLE) by combining FDG-PET and FMZ-PET with statistical feature extraction and machine learning.
This study included 20 drug-resistant patients with unilateral TLE (14 mesial TLE, 6 lateral TLE) and 2 control groups ( = 29 for FDG, = 20 for FMZ). EZ of each patient was confirmed by postsurgical pathology and 1-year follow-up, while propagation zone (PZ) and noninvolved zone (NIZ) were derived from the epileptogenicity index based on presurgical stereo-encephalography (SEEG) monitoring. Whole brain PET scans were obtained with dual tracers [F]FDG and [F]FMZ on separate days, from which standard uptake value ratio (SUVR) was calculated by global mean scaling. Low-order statistical parameters of SUVRs and t-maps derived against control groups were extracted. Additionally, fused FDG and FMZ features were created by using arithmetic operations. Spearman correlation was used to investigate the associations between FDG and FMZ, while multiple linear regression analyses were used to explore the interaction effects of imaging features in predicting epileptogenicity. Crafted imaging features were used to train logistic regression models to predict EZ, whose performance was evaluated by using 10-fold cross-validation at ROI level, and leave-1-patient-out cross-validation at patient level.
FDG SUVR significantly decreased in EZ and PZ compared with NIZ, while FMZ SUVR in EZ significantly differed from PZ. Interaction effects were found between FDG and FMZ in their prediction of epileptogenicity. Fusion of FDG and FMZ provided the best prediction model with an area under the curve (AUC) of 0.86 [0.84-0.87] for EZ versus NIZ and an AUC of 0.79 [0.77-0.81] for EZ versus PZ, eliminating 100% false-positives in 50% of patients, and ≥80% FPs in 90% of patients at patient level.
Combined FDG and FMZ offer a promising avenue for noninvasive localization of the epileptogenic zone in TLE, potentially refining surgical planning.
癫痫是一种全球流行的神经系统疾病，为了进行有效的手术治疗，需要精确识别致痫区（EZ）。虽然已证实氟代脱氧葡萄糖正电子发射断层扫描（FDG-PET）和氟马西尼（FMZ）-PET各自的效用，但它们在定位致痫区方面的联合疗效仍未得到充分探索。我们旨在通过将FDG-PET和FMZ-PET与统计特征提取及机器学习相结合，改善对颞叶癫痫（TLE）中致痫区的无创预测。
本研究纳入了20例单侧TLE的耐药患者（14例内侧TLE，6例外侧TLE）以及2个对照组（FDG组29例，FMZ组20例）。每位患者的致痫区通过术后病理及1年随访得以确认，而传播区（PZ）和非受累区（NIZ）则根据术前立体定向脑电图（SEEG）监测的致痫性指数得出。在不同日期使用双示踪剂[F]FDG和[F]FMZ进行全脑PET扫描，通过全局均值缩放计算标准摄取值比率（SUVR）。提取SUVR及相对于对照组得出的t图的低阶统计参数。此外，通过算术运算创建融合的FDG和FMZ特征。使用Spearman相关性研究FDG与FMZ之间的关联，同时使用多元线性回归分析探索成像特征在预测致痫性方面的相互作用。精心设计的成像特征用于训练逻辑回归模型以预测致痫区，其性能在感兴趣区（ROI）水平通过10倍交叉验证进行评估，在患者水平通过留一患者交叉验证进行评估。
与NIZ相比，EZ和PZ中的FDG SUVR显著降低，而EZ中的FMZ SUVR与PZ有显著差异。在预测致痫性方面发现FDG与FMZ之间存在相互作用。FDG和FMZ的融合提供了最佳预测模型，对于EZ与NIZ，曲线下面积（AUC）为0.86[0.84 - 0.87]，对于EZ与PZ，AUC为0.79[0.77 - 0.81]，在患者水平消除了50%患者中的100%假阳性以及90%患者中的≥80%假阳性。
联合使用FDG和FMZ为TLE中致痫区的无创定位提供了一条有前景的途径，可能会优化手术规划。
