02cor/ 9172
Date set: 22/01/03
Anticipation of epileptic seizures from standard EEG recordings
Sir—2 years ago, M Le Van Quyen and colleagues
1reported anticipation of epileptic seizures from non-linear changes in standard electroencephalograms (EEGs) in 25 of 26 recordings of mesial temporal lobe complex partial seizures (mTLCPS).
We tried to replicate this work and to detect a preictal state preceding mTLCPS using the similarity measure of scalp EEG recordings. We were not able to replicate the findings of Le Van Quyen and colleagues.
We studied a sample of 13 patients who required continuous scalp EEG recording and video monitoring to localise seizure onset. 12 patients had refractory mTLCPS associated with hippocampal sclerosis, and became seizure-free for more than 2 years after surgery. The EEGs were recorded on 22-channel OSG EEG recorders. The electrodes were placed according to the international 10–20 system. The scalp potentials were sampled at 250 Hz. To avoid changes induced by variation of arousal states, only seizures in which the patients were awake for the entire recording were studied. The EEG data for 60 min preceding the clinical seizure onset were used in the analysis. Clinical seizure onset was determined by an epileptologist. The 13th patient had neocortical partial seizures (NPS) and underwent simultaneous scalp and intracranial EEG recordings. The EEG data of this patient were kindly provided to us by Le Van Quyen and colleagues.
2The similarity measure
1was not able to detect preictal changes in the scalp EEG recordings of any of the 12 patients with mTLCPS. However, in both intracranial and scalp EEG recordings of the patient with NPS, the similarity measure identified a preictal state. The detection of a preictal state in the patient with NPS ensured us that the similarity measure was correctly implemented and had the potential to identify a preictal state in scalp EEG recordings.
Possible reasons for the discrepancies between our findings and those reported by Le Van Quyen and colleagues could be that evolution towards a seizure does not always contain a preictal transitional phase, that
technical differences exist between EEG recorders, or that the scalp EEG recordings might need additional preprocessing.
WDC and PL are supported by FWO Research sponsored by the Flemish Government FWO, Belgian Federal DWTC Programme IUAP V-22, and KULeuven research council: GOA-Mefisto 666, IDO /99/003.
W De Clercq, P Lemmerling, S Van Huffel,
*W Van Paesschen
Department of Electrical Engineering, ESAT- SCD(SISTA), Katholieke Universiteit Leuven, Leuven, Belgium; and *Department of Neurology, University Hospital Gasthuisberg, 3000 Leuven, Belgium
(email: wim.declercq@esat.kuleuven.ac.be) 1 Le Van Quyen M, Martinerie J, Navarro V, et
al. Anticipation of epileptic seizures from standard EEG recordings. Lancet 2001; 357:
183–88.
2 Navarro V, Martinerie J, Le Van Quyen M, et al. Seizure anticipation in human neocortical partial epilepsy. Brain 2002; 125: 640–55.
02cor/ 11182
Date set: 22/01/03
follows
Sir—We corresponded with Jacques Martinerie on May 2, 2002, sending details of the program we used (derived from his paper) and the signal to which we applied this program. Therefore, stating that we did not provide them with our data is inaccurate.
In that correspondence, we wrote that we did use the SVD, so it is very strange that they think we did not. Moreover, we asked whether our code contained some errors. In his reply, Martinerie congratulated us on our code and suggested some fine-tuning of the parameters, which we tried without success. Furthermore, one of our main domains of expertise is numerical linear algebra, in which the SVD plays a major role. So if there is one thing we would not forget it would be the SVD.
Wim Van Paesschen is an expert in the domain of epileptology and since reference dynamics were chosen with his help, it is hard to believe that we made a mistake at this stage.
We agree that technical differences
might exist between EEG recorders, but is
not the conclusion or advice that they gave
us. They simply suggested that we do more
measurements so that eventually we would
encounter a signal that allowed for
prediction. In our opinion, this indicates a
method that is not robust and yields many
false negatives.
Contrary to what Le Van Quyen and colleagues suggest, we did apply the algorithm to all channels, with the same conclusion for each: no prediction was possible.
We do not deny the possible use of non- linear methods in medicine; however, we think that although the paper Le Van Quyen and colleagues indicates a promising method by which to analyse EEG methods, its robustness is questionable. McSharry and colleagues
1have shown, by using Le Van Quyen and co-workers’ own data, that a simple measure such as variance was almost as good at predicting seizures as their complicated non-linear approach. We still have doubts about the reproducibility of their results, and suspect that they have many false negatives when all measurements are used, not only those allowing for predictions.
W De Clercq, P Lemmerling, S Van Huffel,
*W Van Paesschen
Department of Electrical Engineering, ESAT- SCD(SISTA), Katholieke Universiteit Leuven, Leuven, Belgium; and *Department of Neurology, University Hospital Gasthuisberg, 3000 Leuven, Belgium
(email: wim.declercq@esat.kuleuven.ac.be) 1 McSharry PE, Smith LA, Tarassenko L.
Prediction of epileptic seizures: are nonlinear methods relevant? Nat Med (in review).
http://www.maths.ox.ac.uk/~mcsharry/papers/
natmededitor.pdf
