Machine learning approach for the outcome prediction of temporal lobe epilepsy surgery

Abstract

Epilepsy surgery is effective in reducing both the number and frequency of seizures, particularly in temporal lobe epilepsy (TLE). Nevertheless, a significant proportion of these patients continue suffering seizures after surgery. Here we used a machine learning approach to predict the outcome of epilepsy surgery based on supervised classification data mining taking into account not only the common clinical variables, but also pathological and neuropsychological evaluations. We have generated models capable of predicting whether a patient with TLE secondary to hippocampal sclerosis will fully recover from epilepsy or not. The machine learning analysis revealed that outcome could be predicted with an estimated accuracy of almost 90% using some clinical and neuropsychological features. Importantly, not all the features were needed to perform the prediction; some of them proved to be irrelevant to the prognosis. Personality style was found to be one of the key features to predict the outcome. Although we examined relatively few cases, findings were verified across all data, showing that the machine learning approach described in the present study may be a powerful method. Since neuropsychological assessment of epileptic patients is a standard protocol in the pre-surgical evaluation, we propose to include these specific psychological tests and machine learning tools to improve the selection of candidates for epilepsy surgery.

Figure 1. Number of times that features were included in the different intermediate subsets selected by the race search feature selection for over 1,000 dataset resamplings.

Figure 2. Estimated classification performance using LOOCV validation.

Only features available before surgery were included in this performance analysis. The x-axis reflects the size of the subset of features retained. A) The upper chart shows the estimated accuracy; whereas, B) the lower chart shows the associated area under the ROC curve. Note that the features for a given point on the x-axis can differ depending on the classifier used (see Table 3 for the respective feature subsets).