Department of Pediatrics, Children's Hospital of Michigan, Wayne State University, Detroit Medical Center, Detroit, MI, 48201, USA.
Department of Pediatrics, Yokohama City University Medical Center, Yokohama, 2320024, Japan.
Sci Rep. 2019 Nov 22;9(1):17385. doi: 10.1038/s41598-019-53749-3.
Statistical parametric mapping (SPM) is a technique with which one can delineate brain activity statistically deviated from the normative mean, and has been commonly employed in noninvasive neuroimaging and EEG studies. Using the concept of SPM, we developed a novel technique for quantification of the statistical deviation of an intracranial electrocorticography (ECoG) measure from the nonepileptic mean. We validated this technique using data previously collected from 123 patients with drug-resistant epilepsy who underwent resective epilepsy surgery. We determined how the measurement of statistical deviation of modulation index (MI) from the non-epileptic mean (rated by z-score) improved the performance of seizure outcome classification model solely based on conventional clinical, seizure onset zone (SOZ), and neuroimaging variables. Here, MI is a summary measure quantifying the strength of in-situ coupling between high-frequency activity at >150 Hz and slow wave at 3-4 Hz. We initially generated a normative MI atlas showing the mean and standard deviation of slow-wave sleep MI of neighboring non-epileptic channels of 47 patients, whose ECoG sampling involved all four lobes. We then calculated 'MI z-score' at each electrode site. SOZ had a greater 'MI z-score' compared to non-SOZ in the remaining 76 patients. Subsequent multivariate logistic regression analysis and receiver operating characteristic analysis to the combined data of all patients revealed that the full regression model incorporating all predictor variables, including SOZ and 'MI z-score', best classified the seizure outcome with sensitivity/specificity of 0.86/0.76. The model excluding 'MI z-score' worsened its sensitivity/specificity to 0.86/0.48. Furthermore, the leave-one-out analysis successfully cross-validated the full regression model. Measurement of statistical deviation of MI from the non-epileptic mean on invasive recording is technically feasible. Our analytical technique can be used to evaluate the utility of ECoG biomarkers in epilepsy presurgical evaluation.
统计参数映射（SPM）是一种可以描绘大脑活动偏离正常均值的技术，已广泛应用于无创神经影像学和脑电图研究中。我们利用 SPM 的概念，开发了一种新的技术，用于量化颅内脑电图（ECoG）测量值与非癫痫均值的统计偏差。我们使用之前从 123 名接受致痫性癫痫手术的耐药性癫痫患者中收集的数据验证了该技术。我们确定了调制指数（MI）从非癫痫均值的统计偏差测量（通过 z 分数评定）如何提高仅基于常规临床、发作起始区（SOZ）和神经影像学变量的癫痫发作结果分类模型的性能。在这里，MI 是一种汇总度量，用于量化 >150 Hz 高频活动与 3-4 Hz 慢波之间原位耦合的强度。我们最初生成了一个规范 MI 图谱，显示了 47 名患者的相邻非癫痫通道的慢波睡眠 MI 的均值和标准差，他们的 ECoG 采样涉及所有四个脑叶。然后，我们计算了每个电极部位的“MI z 分数”。在剩余的 76 名患者中，SOZ 的“MI z 分数”高于非 SOZ。对所有患者的综合数据进行多元逻辑回归分析和接收器操作特性分析表明，包含所有预测变量（包括 SOZ 和“MI z 分数”）的完整回归模型能够以 0.86/0.76 的敏感性/特异性最佳分类癫痫发作结果。排除“MI z 分数”的模型将其敏感性/特异性降低至 0.86/0.48。此外，一次剔除一个病例的分析成功地对完整回归模型进行了交叉验证。从侵入性记录中测量 MI 从非癫痫均值的统计偏差在技术上是可行的。我们的分析技术可用于评估 ECoG 生物标志物在癫痫术前评估中的效用。
