Mario Chávez

Remy Ben Messaoud, Mario Chavez

Epileptic seizure prediction has gained considerable interest in the computational Epilepsy research community. This paper presents a Machine Learning based method for epileptic seizure prediction which outperforms state-of-the art methods. We compute a probability for a given epoch, of being pre-ictal against interictal using the Random Forest classifier and introduce new concepts to enhance the robustness of the algorithm to false alarms. We assessed our method on 20 patients of the benchmark scalp EEG CHB-MIT dataset for a seizure prediction horizon (SPH) of 5 minutes and a seizure occurrence period (SOP) of 30 minutes. Our approach achieves a sensitivity of 82.07 % and a low false positive rate (FPR) of 0.0799 /h. We also tested our approach on intracranial EEG recordings.

X. Navarro, F. Porée, M. Kuchenbuch et al.

The study of electroencephalographic (EEG) bursts in preterm infants provides valuable information about maturation or prognostication after perinatal asphyxia. Over the last two decades, a number of works proposed algorithms to automatically detect EEG bursts in preterm infants, but they were designed for populations under 35 weeks of post menstrual age (PMA). However, as the brain activity evolves rapidly during postnatal life, these solutions might be under-performing with increasing PMA. In this work we focused on preterm infants reaching term ages (PMA $\geq$ 36 weeks) using multi-feature classification on a single EEG channel. Five EEG burst detectors relying on different machine learning approaches were compared: Logistic regression (LR), linear discriminant analysis (LDA), k-nearest neighbors (kNN), support vector machines (SVM) and thresholding (Th). Classifiers were trained by visually labeled EEG recordings from 14 very preterm infants (born after 28 weeks of gestation) with 36 - 41 weeks PMA. The most performing classifiers reached about 95\% accuracy (kNN, SVM and LR) whereas Th obtained 84\%. Compared to human-automatic agreements, LR provided the highest scores (Cohen's kappa = 0.71) and the best computational efficiency using only three EEG features. Applying this classifier in a test database of 21 infants $\geq$ 36 weeks PMA, we show that long EEG bursts and short inter-bust periods are characteristic of infants with the highest PMA and weights. In view of these results, LR-based burst detection could be a suitable tool to study maturation in monitoring or portable devices using a single EEG channel.

X Navarro-Sune, A. L. Hudson, F. De Vico Fallani et al.

During mechanical ventilation, patient-ventilator disharmony is frequently observed and may result in increased breathing effort, compromising the patient's comfort and recovery. This circumstance requires clinical intervention and becomes challenging when verbal communication is difficult. In this work, we propose a brain computer interface (BCI) to automatically and non-invasively detect patient-ventilator disharmony from electroencephalographic (EEG) signals: a brain-ventilator interface (BVI). Our framework exploits the cortical activation provoked by the inspiratory compensation when the subject and the ventilator are desynchronized. Use of a one-class approach and Riemannian geometry of EEG covariance matrices allows effective classification of respiratory states. The BVI is validated on nine healthy subjects that performed different respiratory tasks that mimic a patient-ventilator disharmony. Classification performances, in terms of areas under ROC curves, are significantly improved using EEG signals compared to detection based on air flow. Reduction in the number of electrodes that can achieve discrimination can often be desirable (e.g. for portable BCI systems). By using an iterative channel selection technique, the Common Highest Order Ranking (CHOrRa), we find that a reduced set of electrodes (n=6) can slightly improve for an intra-subject configuration, and it still provides fairly good performances for a general inter-subject setting. Results support the discriminant capacity of our approach to identify anomalous respiratory states, by learning from a training set containing only normal respiratory epochs. The proposed framework opens the door to brain-ventilator interfaces for monitoring patient's breathing comfort and adapting ventilator parameters to patient respiratory needs.