Towards an Online Seizure Advisory System-An Adaptive Seizure Prediction Framework Using Active Learning Heuristics

Abstract

In the last decade, seizure prediction systems have gained a lot of attention because of their enormous potential to largely improve the quality-of-life of the epileptic patients. The accuracy of the prediction algorithms to detect seizure in real-world applications is largely limited because the brain signals are inherently uncertain and affected by various factors, such as environment, age, drug intake, etc., in addition to the internal artefacts that occur during the process of recording the brain signals. To deal with such ambiguity, researchers transitionally use active learning, which selects the ambiguous data to be annotated by an expert and updates the classification model dynamically. However, selecting the particular data from a pool of large ambiguous datasets to be labelled by an expert is still a challenging problem. In this paper, we propose an active learning-based prediction framework that aims to improve the accuracy of the prediction with a minimum number of labelled data. The core technique of our framework is employing the Bernoulli-Gaussian Mixture model (BGMM) to determine the feature samples that have the most ambiguity to be annotated by an expert. By doing so, our approach facilitates expert intervention as well as increasing medical reliability. We evaluate seven different classifiers in terms of the classification time and memory required. An active learning framework built on top of the best performing classifier is evaluated in terms of required annotation effort to achieve a high level of prediction accuracy. The results show that our approach can achieve the same accuracy as a Support Vector Machine (SVM) classifier using only 20 % of the labelled data and also improve the prediction accuracy even under the noisy condition.

Keywords: EEG; epilepsy; health-care; implantable body sensor networks; machine learning; seizure prediction; signal processing.

Figure 1

Example of a schematic flow of closed loop operation of the DBS process. Inclusion of the expert is always in loop before any treatment is applied locally. Picture adapted under CC BY 3.0 license from [20].

Figure 2

Four states of ElectroCorticoGraph (ECoG) of an epileptic patient [22]. A snapshot of data captured from an array of 16 brain implanted electrodes of an epileptic patient with generalized tonic-clonic seizure. The sampling rate is 500 Hz and is recorded for a total duration of 40 h. Duration of the seizure prediction horizon and the seizure occurrence period are illustrated with respect to the onset of a seizure alarm. (A snapshot of ECoG signal obtained from intracranial ElectroEncephaloGraphy (iEEG) viewer [24].)

Figure 3

Schematic flow of our seizure prediction framework. The base classifier is expected to output both soft label (or degree of certainty), which is used to determine the certainty of the classification and crisp label (or label) to know the class.

Figure 4

Schematic flow of active learner block based on Bernoulli-Gaussian model. This model takes the ambiguous samples along with the base classifier’s label prediction with its certainty as inputs.

Figure 5

Performance of classifiers for different sets of features.

Figure 6

Accuracy $\left(\mathcal{A}\right)$ of a linear Support Vector Machine classifier as a function of window sizes.

Figure 7

Classification time for 20 s of Electro Cortico Graphy data from 16 channels using $se{t}_t$ feature set, timing with Dog5 data.

Figure 8

Number of seizure episodes missed from classification as a function of threshold for all dataset.

Figure 9

Label complexity of three different settings of the Support Vector Machine classifier. In each setting, the error rate is measured as a function of label fraction.

Figure 10

Label complexity of three different settings of the Support Vector Machine classifier. Error rate as a function of label fraction.

Figure 11

Noise complexity of three different settings of the the Support Vector Machine classifier. Accuracy as a function of noise rate $\left(\eta \right)$.