Detection and Prediction of Epileptic Seizures

Jonas Duun-Henriksen

    Research output: Book/ReportPh.D. thesis

    1631 Downloads (Pure)


    Approximately 50 million people worldwide suffer from epilepsy. Although 70% can control their seizures by anti-epileptic drugs, it is still a cumbersome disease to live with for a large group of patients. The current PhD dissertation investigates how these people can be helped by continous monitoring of their brain waves. More specifically, three issues were investigated: The feasibility of automatic seizure prediction, optimization of automatic seizure detection algorithms, and the link between intra- and extracranial EEG. Regarding feasibility of automatic seizure prediction, neither the author nor any other in the seizure prediction society have yet obtained clinical applicable prediction results. However, this should not be taken as discouraging for the future. New large public databases have emerged during 2012 which might provide the means to identify patterns leading to reliable seizure prediction algorithms.
    More promising results were obtained in the investigating of possible use of an outpatient EEG monitoring device for idiopathic generalized epilepsy patients. Combined with an automatic seizure detection algorithm such a device can give an objective account of the paroxysm frequency, duration, and time of occurrence. Based on standard EEG data from 20 patients recorded in the clinic, the log-sum of wavelet transform coefficients were used as feature input to a classifier consisting of a support vector machine. 97% of paroxysms lasting more than two seconds were correctly detected without any false positive detections. This was obtained using a generic algorithm on the signals from only a single frontal channel. Applying the same algorithm architecture on EEG data from two outpatient children monitored for approximately three entire days each, the sensitivity was 90% and the false detection rate was 0.12/h. When more recordings are collected, the outpatient algorithm can be further optimized and results should improve.
    The final investigation examined the relationship between spontaneous, awake intra- and extracranial EEG. Seven patients with electrodes placed subdurally as well as subgaleally were used to estimate the field of vision of a single extracranial channel. By computation of the coherence between the channels, the well recognized hypothesis stating that the skull acts as an electroencephalographic averager was proven correct. Although coherence was significant in an accumulation area of 150 cm2, only channels within a cortical area of approximately 30 cm2 showed to increase the coherence. The increase seemed to progress linearly with an accumulation area up to 31 cm2, where 50% of the maximal coherence was accumulated from only 2 cm2 (corresponding to one channel), and 75% from 16 cm2. The coherences of different frequency bands below 16 Hz all seem to have similar declines as a function of the Euclidean distance between channels. Frequencies between 16 and 30 Hz have a steeper decline and will only show coherent parts to cortical channels within 60 cm2. There is no coherence for frequencies above 30 Hz at any distance.
    A lot of patients with epilepsy still struggle with a dreadful fear of suddenly having a seizure. The current PhD study identified topics where an EEG monitor could provide improvement in the patient’s quality of life. By algorithm development, implementation and testing, a step toward such a device is presented.

    Original languageEnglish
    Place of PublicationKgs. Lyngby
    PublisherTechnical University of Denmark
    Number of pages140
    ISBN (Print)978-87-92465-62-7
    Publication statusPublished - 2013


    Dive into the research topics of 'Detection and Prediction of Epileptic Seizures'. Together they form a unique fingerprint.

    Cite this