Malformations of cortical development (MCD) comprise a broad spectrum of structural brain lesions frequently associated with epilepsy. Disease definition and diagnosis remain challenging and are often prone to arbitrary judgment. Molecular classification of histopathological entities may help rationalize the diagnostic process. We present a retrospective, multi-center analysis of genome-wide DNA methylation from human brain specimens obtained from epilepsy surgery using EPIC 850 K BeadChip arrays. A total of 308 samples were included in the study. In the reference cohort, 239 formalin-fixed and paraffin-embedded (FFPE) tissue samples were histopathologically classified as MCD, including 12 major subtype pathologies. They were compared to 15 FFPE samples from surgical non-MCD cortices and 11 FFPE samples from post-mortem non-epilepsy controls. We applied three different statistical approaches to decipher the DNA methylation pattern of histopathological MCD entities, i.e., pairwise comparison, machine learning, and deep learning algorithms. Our deep learning model, which represented a shallow neuronal network, achieved the highest level of accuracy. A test cohort of 43 independent surgical samples from different epilepsy centers was used to test the precision of our DNA methylation-based MCD classifier. All samples from the test cohort were accurately assigned to their disease classes by the algorithm. These data demonstrate DNA methylation-based MCD classification suitability across major histopathological entities amenable to epilepsy surgery and age groups and will help establish an integrated diagnostic classification scheme for epilepsy-associated MCD.

皮质发育畸形 (MCD) 包括经常与癫痫相关的广泛的结构性脑损伤。疾病定义和诊断仍然具有挑战性，并且往往容易被武断判断。组织病理学实体的分子分类可能有助于使诊断过程合理化。我们使用 EPIC 850 K BeadChip 阵列对从癫痫手术中获得的人脑样本的全基因组 DNA 甲基化进行回顾性、多中心分析。共有 308 个样本被纳入研究。在参考队列中，239 个福尔马林固定和石蜡包埋 (FFPE) 组织样本在组织病理学上被归类为 MCD，包括 12 个主要亚型病理。将它们与来自外科非 MCD 皮质的 15 个 FFPE 样本和来自死后非癫痫对照的 11 个 FFPE 样本进行比较。我们应用了三种不同的统计方法来破译组织病理学 MCD 实体的 DNA 甲基化模式，即成对比较、机器学习和深度学习算法。我们的深度学习模型代表了一个浅层神经元网络，达到了最高水平的准确度。来自不同癫痫中心的 43 个独立手术样本的测试队列用于测试我们基于 DNA 甲基化的 MCD 分类器的精度。来自测试队列的所有样本都被算法准确地分配到他们的疾病类别中。这些数据表明基于 DNA 甲基化的 MCD 分类适用于适合癫痫手术和年龄组的主要组织病理学实体，并将有助于建立癫痫相关 MCD 的综合诊断分类方案。