Transfer-Learning for Differentiating Epileptic Patients Who Respond to Treatment Based on Chronic Ambulatory ECoG Data

Abstract

The aim of this study was to evaluate whether transfer-learning with pre-trained deep convolutional neural networks (deep CNNs) can be used for assessing patient outcomes in epilepsy. Transfer-learning with the GoogLeNet InceptionV3 CNN model pre-trained on the large ImageNet dataset (~1.2 million images) was able to differentiate upper (n=12) and lower (n=9) response quartile mesiotemporal lobe epilepsy patients in the NeuroPace® RNS® System clinical trials with ~76% classification accuracy based on chronic ambulatory baseline electrocorticographic (ECoG) data. These promising findings justify further research using deep CNNs for assessing patient outcomes in epilepsy.