A Long Short-Term Memory neural network for the detection of epileptiform spikes and high frequency oscillations.

Over the last two decades, the evidence has been growing that in addition to epileptic spikes high frequency oscillations (HFOs) are important biomarkers of epileptogenic tissue. New methods of artificial intelligence such as deep learning neural networks can provide additional tools for automated analysis of EEG. Here we present a Long Short-Term Memory neural network for detection of spikes, ripples and ripples-on-spikes (RonS). We used intracranial EEG (iEEG) from two independent datasets. First dataset (7 patients) was used for network training and testing. The second dataset (5 patients) was used for cross-institutional validation. 1000 events of each class (spike, RonS, ripple and baseline) were selected from the candidates initially found using a novel threshold method. Network training was performed using random selections of 50-500 events (per class) from all patients from the 1st dataset. This 'global' network was then tested on other events for each patient from both datasets. The network was able to detect events with a good generalisability namely, with total accuracy and specificity for each class exceeding 90% in all cases, and sensitivity less than 86% in only two cases (82.5% for spikes in one patient and 81.9% for ripples in another patient). The deep learning networks can significantly accelerate the analysis of iEEG data and increase their diagnostic value which may improve surgical outcome in patients with localization-related intractable epilepsy.

Pathophysiology of altered consciousness during seizures: Subtraction SPECT study.

BACKGROUND: The mechanisms underlying altered consciousness during seizures are poorly understood. Previous clinicopathologic studies suggest a role for the thalamus and upper brainstem in consciousness mechanisms. OBJECTIVE: To examine blood flow changes associated with altered consciousness during seizures. METHODS: Seventy-one patients with epilepsy who underwent video-EEG monitoring and ictal/interictal SPECT were studied. Patients were divided into three groups depending on their conscious state during seizures: 1) complete impairment of consciousness (CI), 2) no impairment of consciousness (NI), or 3) uncertain impairment of consciousness (UI). The distribution of blood flow changes during these seizures was assessed by subtraction (ictal - interictal) SPECT co-registered to MRI. Conscious state was assessed in relation to secondary ictal hyperperfusion in subcortical regions (i.e., thalamus and upper brainstem). RESULTS: Impairment of consciousness showed a strong association with secondary hyperperfusion in the thalamic/upper brainstem region (p = 0.01), occurring in 92% (45/49) of CI, 69% (9/13) of UI, and 11% (1/9) of NI. CONCLUSIONS: These findings are consistent with a role for the thalamus and upper brainstem in consciousness mechanisms. The authors suggest that the spread of epileptic discharges or a trans-synaptic activation (diaschisis) of these structures is an important mechanism in the alteration of consciousness during seizures. Variance in the results may be due to differences in timing of radioisotope injection, sensitivity of the subtraction SPECT technique, and the ability to clinically assess the conscious state.

Differential cognitive and behavioral effects of carbamazepine and lamotrigine.

BACKGROUND: The relative cognitive and behavioral effects of lamotrigine compared with the older standard antiepileptic drugs (AED) are uncertain. OBJECTIVE: To directly compare the cognitive and behavioral effects of carbamazepine and lamotrigine. METHODS: The cognitive and behavioral effects of carbamazepine and lamotrigine were assessed in 25 healthy adults using a double-blind, randomized crossover design with two 10-week treatment periods. During each treatment condition, subjects received either lamotrigine (150 mg/day) or carbamazepine (mean 696 mg/day) adjusted to a dose to achieve midrange standard therapeutic blood levels (mean 7.6 microg/mL). Subjects were tested at the end of each AED treatment period and in three drug-free conditions (two pretreatment baselines and a final posttreatment period [1 month after last AED]). The neuropsychological test battery included 19 measures yielding 40 total variables. RESULTS: Direct comparison of the two AED revealed significantly better performance on 19 (48%) variables for lamotrigine but none for carbamazepine. Differences spanned both objective cognitive and subjective behavioral measures, including cognitive speed, memory, graphomotor coding, neurotoxic symptoms, mood factors, sedation, perception of cognitive performance, and other quality-of-life perceptions. Comparison of carbamazepine with the nondrug average revealed significantly better performance for nondrug average on 24 (62%) variables but none for carbamazepine. Comparison of lamotrigine with nondrug average revealed better performance on one (2.5%) variable for nondrug average and on one (2.5%) variable for lamotrigine. CONCLUSION: Lamotrigine produces significantly fewer untoward cognitive and behavioral effects than carbamazepine at the dosages used in this study.

Prognostic implication of contralateral secondary electrographic seizures in temporal lobe epilepsy.

PURPOSE: Interhemispheric propagation of seizures in temporal lobe epilepsy is frequently noted during intracranial EEG monitoring. We hypothesized that a distinct secondary electrographic seizure (DSES) in the temporal lobe contralateral to primary seizure onset may be an unfavorable prognostic indicator. METHODS: We reviewed intracranial depth electrode EEG recordings, 1-year outcome, and medical records of 51 patients (M 29, F 22: age 15-64 years) who underwent anterior temporal lobectomy during 1988-96. We defined DSES as a seizure that spread to the contralateral temporal lobe and produced distinct contralateral EEG features. The distinct feature was focal involvement of one or two electrode contacts at onset, which starts and evolves independently from the ipsilateral temporal lobe. We considered DSES as the predominant seizure pattern when it occurred in more than one half of the patients' recorded seizures. RESULTS: Only nine of 19 (47%) patients with predominant DSES had a 1-year seizure-free outcome, whereas 27 of 32 (84%) patients without predominant DSES had a 1-year seizure-free outcome (p < 0.01). Bitemporal independent seizures were more common in patients with predominant DSES (9/19 versus 0/32; p < 0.001). CONCLUSION: Our results suggest that distinct contralateral secondary electrographic seizure is a predictor of unfavorable outcome and is also more likely to be associated with bitemporal seizures.

Differential cognitive effects of carbamazepine and gabapentin.

PURPOSE: The cognitive effects of the newer antiepileptic drugs (AEDs) compared with the older standard AEDs are uncertain. METHODS: We directly compared the cognitive effects of carbamazepine (CBZ) and gabapentin (GBP) in 35 healthy subjects by using a double-blind, randomized crossover design with two 5-week treatment periods. During each treatment condition, subjects received either GBP, 2,400 mg/day, or CBZ (mean, 731 mg/day) adjusted to a dose to achieve midrange standard therapeutic blood levels (mean, 8.3 microg/ml). Subjects were tested at the end of each AED treatment period and in four drug-free conditions [two pretreatment baselines and two post-treatment washout periods (1 month after each AED)]. The neuropsychological test battery included 17 measures yielding 31 total variables. RESULTS: Direct comparison of the two AEDs revealed significantly better performance on eight variables for GBP, but none for CBZ. Comparison of CBZ and GBP to the nondrug average revealed significant statistical differences for 15 (48%) of 31 the variables. Pairwise follow-up analyses of the 15 variables revealed significantly better performance for nondrug average on 13 variables compared with CBZ, and on four compared with GBP. GBP was better than nondrug average on one variable. CONCLUSIONS: Although both CBZ and GBP produced some effects, GBP produced significantly fewer untoward cognitive effects compared with CBZ at the dosages used in this study.

A 13-year experience with epilepsy surgery.

Between 1985 and 1997, 563 therapeutic craniotomies were performed: 311 anterior temporal (ATL) and 158 extramesial temporal (XMT) resections, 67 callosotomies, 20 hemispherectomies and 7 multiple subpial transections. Sixty-seven percent of nonlesional ATL cases were seizure free (SF), and 76% of nonlesional ATL cases < or =18 years old were SF. Seventy-eight percent of lesional ATL cases with complete resection were SF. Seventy-three percent of lesional cases < or =18 were SF. Thirty-seven percent of nonlesional XMT cases were SF. Seventy percent of XMT lesional cases with complete resection were SF, and 82% of lesional XMT cases < or =18 were SF. Of the anterior callosotomy cases, there was a > or =90% decrease in generalized tonic-clonic seizures in 50% of patients, and in tonic seizures, drop attacks, absence and myoclonic seizures in approximately 60-70% of patients. Of 20 hemispherectomies, 65% were SF. Of 7 multiple subpial transections, 29% were SF.

Wada memory and timing of stimulus presentation.

We compared Wada memory performance for stimuli presented at two timing intervals following amobarbital injection in 47 non-lesional patients with complex partial seizures (L = 26; R = 21). A significant interaction between seizure focus and timing of presentation was present (P < 0.03). Memory performance for objects whose presentation began approximately 50-55 s following amobarbital administration differed as a function of ipsilateral vs. contralateral injection at a very high level of statistical significance (P < 0.00001). Items presented approximately 4 min, 30 s post injection were also related to seizure onset literality, but at a lower statistical level (P < 0.01). Presentation of Wada memory stimuli earlier during hemispheric anaesthesia yields results that are more sensitive to lateralized temporal lobe seizure onset than does presentation of items later during the procedure.

Results of lesional vs. nonlesional frontal lobe epilepsy surgery.

Fifty-three seizure focus resections limited to the frontal lobe were performed. Forty-nine had adequate follow-up. Five of 17 (29%) nonlesional (NL) cases and 21 of 32 (66%) lesional (L) cases were seizure free (SF) at 1 year postoperatively. Eight of 9 (89%) L cases < or = 18 years old vs. 13 of 23 (57%) > 18 years old were SF. Eight of 10 (80%) tumor vs. 13 of 22 (59%) nontumor L cases were SF. Sixteen NL cases were localized by invasive recording. Five (31%) were SF. Ictal localization was obtained in 4 of 32 L cases. Three of 4 (75%) were SF. Eighteen of 24 (64%) L cases without ictal localization were SF. Nine of 12 (75%) lateral resections, 7 of 12 (58%) lobectomies, and 2 of 13 (15%) mesial resections were SF. Three of 20 cases with at least 90% reduction in seizures became SD > or = 2 years postresection. No case with < 90% seizure reduction at one year showed improvement with longer follow-up.

The significance of ictal depth EEG patterns in patients with temporal lobe epilepsy.

We reviewed 187 depth recorded seizures in 33 patients with non-lesional temporal lobe complex partial seizures. All patients had a minimum of 1 year follow-up following temporal lobectomy. We classified seizure onset pattern as rhythmic activity, attenuation, or repetitive spikes or spike wave complexes. The most common pattern of seizure onset was rhythmic activity and the next most common pattern was repetitive spikes. Seventy-five seizures (49%) had only one seizure onset pattern, and 79 seizures (51%) had a combination of seizure onset patterns. The degree of hippocampal gliosis strongly predicted the type of seizure onset pattern (Chi square = 24.07, 2 d.f., P < 0.01). The rhythmic activity pattern was associated with mild gliosis, and the repetitive spike pattern was associated with severe gliosis. We classified seizure onset as focal or regional based on the number of electrode contacts that were involved by the ictal EEG. A focal seizure onset was associated with an excellent outcome following temporal lobectomy.

Amobarbital evaluation of neurobehavioral function prior to therapeutic occlusion of brain arteriovenous malformations: a new neuropsychological procedure.

Because untreated arteriovenous malformations (AVMs) frequently result in some form of permanent neurological complication, treatment of AVMs is aggressively pursued A relatively new treatment consists of sending micropellets into blood vessels supplying the AVM core to block blood flow and "shrink" the AVM When vessels supplying the AVM are thought to also irrigate vital portions of brain, evaluations of neurobehavioral function after injection of amobarbital into intracranial vessels (Wada testing) may be performed to prevent significant complications folIowing embolization This study details our preliminary experience with Wada testing and electroencephalography (EEG) prior to AVM embolization in seven patients Neurobehavioral functions were continuously monitored after injection of 50-75 mg of amobarbital into target cerebral vessels No change in sensorimotor, cognitive, or EEG functions were detected in any of the superselective Wada examinations Embolization was performed following all negative Wada evaluations The only irreversible complication after embolization was a superior quadrantanopia No other permanent neurobehavioral sequelae resulted from embolization These preliminary findings suggest that simultaneous Wada/EEG monitoring may be useful in predicting neurobehavioral complications prior to AVM embolization.

Wada memory asymmetries predict verbal memory decline after anterior temporal lobectomy.

We examined Wada memory and neuropsychological memory function in 34 nonlesional patients who underwent anterior temporal lobectomy (ATL) and who were seizure free at 1-year follow-up. Patients who displayed a decline on verbal memory measures that exceeded 1 SD after left ATL had significantly smaller left/right Wada memory asymmetries than left ATL patients without a significant verbal memory decline. When Wada memory asymmetries were used to predict verbal memory decline after left ATL in individual patients, similar statistically significant effects were present. No significant relationship between Wada memory and postoperative memory was present in right ATL patients, and postoperative memory function was not related to Wada memory performance after either left hemisphere or right hemisphere injection alone. We conclude that Wada memory asymmetries provide one measure of the risk to material-specific decline in verbal memory after left ATL.

Precision of dipole localization in a spherical volume conductor: a comparison of referential EEG, magnetoencephalography and scalp current density methods.

In this study, we determined the influence of dipole orientation, dipole location, and number of recording sites on the precision of dipole localization in a spherical volume conductor. We compared localization from referential EEG (R-EEG), scalp current density EEG (SCD-EEG) and magnetoencephalography (MEG). Dipole orientation had a small influence on the precision of dipole localization for R-EEG and SCD-EEG. Dipole location relative to the recording sites, dipole depth, and number of recording channels strongly influenced the precision of dipole localization. Assuming equal signal to noise conditions for each recording method, MEG and SCD-EEG had a similar precision for dipole localization of a single tangential dipole source and both methods were more precise than R-EEG. However, SCD-EEG was inferior to MEG for distinguishing a single tangential current source from a pair of deeper radial current sources. In summary, these results suggest that the MEG will be most useful for localization of multiple simultaneous dipole sources.

Multichannel magnetoencephalography in ablative seizure surgery outside the anteromesial temporal lobe.

Magnetoencephalography (MEG) was used to evaluate 40 candidates for seizure surgery thought to have foci outside the anteromesial temporal lobe. Of 29 cases with electrographic data suggesting a convexity focus, MEG spikes were recorded from 28. In 21, MEG and electrographic data were localized to the same area. Invasive studies were, or could have been, avoided in 13 cases based on MEG and other noninvasive data. MEG was not localizing value in 4 orbitofrontal or 7-depth-nonlocalized cases. Seventeen patients with MEG epileptiform data have had postoperative follow-up. Eight of 13 with electrographic and MEG data localized to the same area are seizure free. None of 4 with spatial discordance of MEG electrographic data are seizure free.

Wada memory performance predicts seizure outcome following anterior temporal lobectomy.

We examined the ability of Wada memory testing to predict seizure outcome 1 year following anterior temporal lobectomy. Asymmetry scores for Wada memory performance, using amobarbital doses of 125 mg or less, were calculated for 55 patients under the age of 45 years who had no radiologic evidence of structural lesions other than gliosis. Wada memory asymmetries were significantly greater (p < 0.02) in patients who were seizure free compared with those who continued to experience seizures. Furthermore, patients with Wada memory score asymmetries of at least three objects (maximum = 8) were more likely to be seizure free compared with patients with Wada memory asymmetries less than three (p < 0.01). Of the 36 patients who had Wada memory score asymmetries of at least three objects, 32 (89%) were seizure free. In contrast, of the 19 patients whose Wada memory score asymmetries were less than three, only 12 (63%) were seizure free. These data suggest that Wada memory performance is related to seizure outcome following anterior temporal lobectomy.

A method for imaging of intracranial EEG electrodes using magnetic resonance imaging.

In this study, we describe a method for imaging intracerebral electrodes within a three-dimensional reconstructed image of the brain. A three-dimensional image of the brain was reconstructed from serial magnetic resonance images. The locations of intracerebral electrodes were determined from anterior-posterior and lateral skull X-rays performed after intracerebral electrode implantation. The three-dimensional reconstruction of the brain including electrode locations was displayed using IRIS Explorer Software (Silicon Graphics, Mountainview, CA). This method might improve the interpretation of electrical patterns of seizure activity recorded from intracerebral electrodes.

Do dichotic word listening asymmetries predict side of temporal lobe seizure onset?

Dichotic word listening asymmetries are thought to be useful in predicting side of temporal lobe dysfunction. However, little direct evidence exists to support this assumption in practice, especially in patients with subtle epileptogenic lesions. To determine if word listening ear asymmetries are valid predictors of side of temporal lobe seizure onset, we examined the preoperative dichotic word listening performance of 80 patients with either left (N = 41) or right (N = 39) temporal lobe (TL) seizure foci. On a group level, patients showed a statistically significant 'lesion effect' as evidenced by a relative deficit in the ear contralateral to the side of lesion. Prediction of side of seizure focus in individual cases, however, was poor: depending upon the criteria used, 61% to 80% of epileptics with unilateral temporal lobe foci did not show the expected contralateral ear deficit. Results suggest that caution be exercised when inferring side of temporal lobe seizure focus through dichotic word listening asymmetries in individual temporal lobe epilepsy cases.

Stimulus timing effects on Wada memory testing.

OBJECTIVE: To determine the effects of presenting Wada memory stimuli at different times after intracarotid amobarbital injection on Wada memory asymmetries. DESIGN: Wada memory asymmetries from three timing series were related to the laterality of eventual temporal lobectomy. SETTING: Academic institution epilepsy surgery program. PATIENTS: Forty-three patients with complex partial seizures who later underwent anterior temporal lobectomy (left temporal lobectomy, 24 patients; right temporal lobectomy, 19 patients). No patient included had abnormalities on magnetic resonance imaging scans to suggest a lesion other than gliosis. RESULTS: Memory performance for objects whose presentation began approximately 45 seconds after amobarbital administration differentiated laterality of seizure onset. Memory for items presented later and after partial return of language (on average 3 minutes 40 seconds postinjection) also differed as a function of ipsilateral vs contralateral injection, but at a lower level of statistical significance. Memory for items presented last during the procedure (on average 6 minutes postinjection) discriminated seizure groups at a still lower level of statistical significance. When used to predict lateralized temporal lobe impairment in individual patients, early object memory performance was significantly better than memory performance employing either middle (56%) or late (43%) stimulus presentation timings. CONCLUSION: The results of early object memory testing are superior to those obtained from stimulus presentation later in the procedure in documenting temporal lobe dysfunction associated with a lateralized seizure onset.

Amnesia after unilateral temporal lobectomy: a case report.

We report a mixed handed (L > R) patient with exclusive right cerebral language representation who developed a permanent anterograde amnestic syndrome after right anterotemporal lobectomy. Preoperative neuropsychological performance consisted of impaired verbal memory and normal nonverbal memory. Wada memory performance was asymmetrical for objects presented soon after amobarbital injection in conjunction with no memory asymmetry for items presented later in the Wada evaluation. Pre- and postoperative magnetic resonance imaging (MRI) scans showed no structural lesions; however, postoperative MRI hippocampal volume measurements suggested decreased hippocampal volume for the nonresected temporal lobe. These results confirm the risk of anterograde amnesia after unilateral temporal lobectomy and demonstrate that baseline neuropsychological testing may falsely literalize material-specific memory functions in patients with atypical cerebral language dominance.

Use of scalp-sphenoidal EEG for seizure localization in temporal lobe epilepsy.

We determined the accuracy and sensitivity of scalp-sphenoidal EEG for seizure focus localization in 50 patients who became seizure-free or had rare seizures following temporal lobectomy. EEG localization was based on concordant interpretations of scalp-sphenoidal ictal EEG by three independent interpreters. All patients became seizure-free or had rare seizures following temporal lobectomy. Localization from EEG disagreed with the side of surgery in only 1 (2%) of 50 patients. We identified 3 distinct patient groups with a low, moderate, and high likelihood of having a focal ictal EEG pattern during a seizure. These groups comprised 31% (low likelihood), 44% (moderate likelihood), and 25% (high likelihood) of patients. A model based on these results suggests that with multiple ictal EEG recordings, accurate localization from scalp-sphenoidal EEG is possible in approximately up to 65-70% of patients with temporal lobe epilepsy.

Role of multichannel magnetoencephalography in the evaluation of ablative seizure surgery candidates.

Magnetic source imaging (MSI) was performed on 30 ablative epilepsy surgery candidates. The technique involved high resolution multiplanar MRI images with lipid fiducials attached to the patient's head to define a head-based 3D coordinate space. Magnetoencephalography (MEG) was performed after digitizing the same fiducial points. A 37-channel magnetometer obtained data at two to five sites over each hemisphere. MEG epileptiform data were obtained with and without EEG triggering. A single equivalent current dipole model was used to determine orientation and location of a dipole generator whose surface isocontour map most closely fits the measured dipolar data for each event. The MEG data were then transformed to MRI images for source localization. In five of seven cases with ictal anterior temporal lobe foci, MSI data were localized to the same temporal lobe but did not provide additional spatial data. In 10 of 11 cases with convexity foci, MSI provided additional spatial localizing data. MSI did not verify depth electrode localization in one anterior temporal-orbital frontal and three orbitofrontal cases. In seven of eight cases in which depth EEG recordings were nonlocalizing, MSI provided insufficient localizing data. MSI appears to provide additional spatial localizing data in most cases with a convexity epileptic focus.