Machine Learning to Classify Relative Seizure Frequency From Chronic Electrocorticography.
Machine Learning to Classify Relative Seizure Frequency From Chronic Electrocorticography.
复制标题
DOI：
10.1097/wnp.0000000000000858
复制
发表时间：
2023-02-01
期刊：
影响因子：
--
通讯作者：
Liu A
中科院分区：
文献类型：
--
作者：
Sun Y;Friedman D;Dugan P;Holmes M;Wu X;Liu A
脑反应性神经刺激（RNS，NeuroPace）治疗难治性局灶性癫痫患者，并提供慢性皮质脑电图（ECoG）。我们探讨了应用于发作间期ECoG的机器学习算法如何评估对神经刺激参数变化的临床反应。我们确定了5例RNS患者，每例患者均具有≥ 200天的连续稳定药物和检测设置（每例患者的中位数为358天）。对于每例患者，每个月的发作间期ECoG片段标记为“高”或“低”，以表示与每月长时间发作（LE）计数中位数相比相对较高或较低的LE计数。从四个RNS通道的六个常规频带的功率被提取作为特征。对于每位患者，五种机器学习算法在80%的ECoG上进行训练，然后在剩余的20%上进行测试。分类器通过受试者工作特征曲线下面积（ROC-AUC）进行评分。我们探讨了癫痫活动的个人昼夜节律周期如何为分类器构建提供信息。支持向量机（SVM）或梯度提升模型实现了最佳性能，范围从0.705（一般）到0.892（优秀）。高伽马功率是最重要的特征，在低频时期趋于降低。在昼夜节律发作高峰期间记录的ECoG训练导致可比较的模型性能，尽管使用的数据较少。对回顾性背景ECoG的机器学习分析可以对个体患者的相对癫痫发作频率进行分类。高伽马功率是最有信息的，而癫痫发作活动的个人昼夜节律模式可以指导模型的建立。建立在发作间期ECOG上的机器学习分类器可以指导刺激编程。
Brain responsive neurostimulation (RNS, NeuroPace) treats patients with refractory focal epilepsy and provides chronic electrocorticography (ECoG). We explored how machine learning algorithms applied to interictal ECoG could assess clinical response to changes in neurostimulation parameters. We identified five RNS patients each with ≥ 200 continuous days of stable medication and detection settings (median 358 days per patient). For each patient, interictal ECoG segments for each month were labeled as “high” or “low” to represent relatively high or low Long Episode (LE) count compared to the median monthly LE count. Power from six conventional frequency bands from four RNS channels were extracted as features. For each patient, five machine learning algorithms were trained on 80% of ECoG then tested on the remaining 20%. Classifiers were scored by the Area under the Receiver Operating Characteristic Curve (ROC-AUC). We explored how individual circadian cycles of seizure activity could inform classifier building. Support Vector Machine (SVM) or Gradient Boosting models achieved the best performance, ranging from 0.705 (fair) to 0.892 (excellent) across patients. High gamma power was the most important feature, tending to decrease during low frequency epochs. Training on ECoG recorded during the circadian ictal peak resulted in comparable model performance, despite less data used. Machine learning analysis on retrospective background ECoG can classify relative seizure frequency for an individual patient. High gamma power was most informative, while individual circadian patterns of seizure activity can guide model building. Machine learning classifiers built on interictal ECOG may guide stimulation programming.