About ENIGMA-Epilepsy  |  Project 1: Gray Matter  |  Project 2: DTI Secondary Projects Protocols  |  Join ENIGMA-Epilepsy  |  Meetings & Minutes FAQ

enigma-epilepsy_logo_cropped

There are currently four ongoing secondary projects, building upon ENIGMA-Epilepsy’s initial investigations of grey and white matter microstructure. Each project, and its respective co-leads, is detailed below.

Secondary Project 1: Gene co-expression

Co-leads: Andre Altmann (UCL), Mina Ryten (UCL), Juan Botía (UCL), Annamaria Vezzani (Istituto Mario Negri), Sanjay Sisodiya (UCL)

Our initial study of grey matter differences in the common epilepsies identified a series of subtle cortical abnormalities across a number of epilepsy subtypes. To explore molecular mechanisms underlying these patterns of regional thickness differences, we sought evidence for genetically-driven regionalised cortical vulnerability by exploring gene expression in the healthy human brain (uncomplicated by secondary changes), based on the neuroanatomically-precise Allen Institute for Brain Science (AIBS) microarray samples. These data were complemented by neuropathological data and data from animal models of epilepsy. Our findings are currently being summarized into a manuscript for peer review.

 

Secondary Project 2: Deep learning

Co-leads: Alicia Chang & Carrie McDonald (UCSD)

Introduction:

EEG video telemetry is the primary clinical procedure used to lateralize and localize seizure foci in patients with epilepsy.  However, this procedure is very time-consuming and often invasive. Therefore, methods that can help to lateralize a seizure focus or provide further confidence of seizure lateralization in a non-invasive, rapid manner are greatly needed.  Deep learning is a novel type of machine learning that is capable of constructing and training neural networks in multi-dimensional datasets, which can be used to answer complex clinical questions.  Unlike older machine learning techniques whose performance was known to plateau, the performance of deep learning is scalable and continues to increase with increasing amounts of data and larger models.  In recent years, this technique has been successfully applied to medical imaging analysis to improve diagnosis (e.g., Alzheimer’s Disease1,2, schizophrenia3), predict survive in amyotropic lateral sclerosis4, as well as to predict “brain age” in neurological patients and healthy controls5. Although the idea behind deep learning is not new, it has only recently become feasible to accomplish in clinical settings due to the availability of greater computational power and data volume. With the availability of improved computer resources and large sample sizes obtained through big data initiatives (e.g., ENIGMA), we are now well-positioned to take our research to the next level by introducing this algorithm to enhance the prediction of seizure lateralization in patients with focal epilepsy. We propose to employ deep learning algorithms, in particular convolutional neural networks, to lateralize the seizure focus in patients with temporal lobe epilepsy (TLE) using T1-weighted structural imaging (e.g., gray matter and white matter) and diffusion tensor imaging (DTI) scans. This proposed project will be achieved by using Tensorflow, which is an open-source software for deep learning6. This software has been designated general enough to be applicable across a wide range of datasets, including those from medical imaging analysis.

In this proposal, we aim to use two approaches to lateralize the seizure focus. First, we propose to use the existing, locally-processed T1-weighted MRI data and available clinical information to lateralize the seizure focus in patients with lesional and non-lesional TLE. The brain measures of interest will be subcortical volumes and cortical thickness estimates derived from gyral-based ROIs. The clinical features will be age, age of seizure onset, disease duration, and MTS status. Using this approach, we aim to compare the classifier performance (i.e., accuracy of prediction) of a hierarchical multi-layer structure algorithm (e.g., deep learning) with a linear shallow structure algorithm (e.g., support vector machine) to determine whether deep learning can provide a more sophisticated feature detection to improve the accuracy of lateralization prediction in the patients with TLE.  With the limited sample size, careful monitoring and feature selection will be employed during the data preparation process. Second, we propose to test the increased accuracy of our model by including both structural and DTI data and a more comprehensive set of clinical features. Since tissue structure and microstructure provide unique information for detecting brain pathology in patients with epilepsy, the combination and integration of these two brain measures combined with relevant clinical characteristics may increase the accuracy of seizure lateralization.

In both approaches, the ground true for each patient will be the lateralization of the seizure focus based on the EEG.  When available, we will further validate our results with post-surgical outcome data.  The collected data will be randomly divided into two parts: (1) training: 80% of the data will be used to optimize the network weights with our deep learning algorithm and (2) testing: the prediction accuracy will based on the remained 20% of the data.

Hypotheses:

Leveraging patients’ subcortical volumes, cortical thickness, DTI data, and clinical features, we hypothesize that we will be able to accurately lateralize the seizure focus in patients with left or right TLE. Although we believe that our predictive ability will be higher for patients with MTS, we believe that we will also obtain high accuracy for seizure lateralization in patients without MTS.

References:

[1] Sarraf, S., & Tofighi, G. (2016). Classification of Alzheimer’s disease structural MRI data by deep learning convolutional neural networks. arXiv preprint arXiv:1607.06583.

[2] Liu, S., Liu, S., Cai, W., Pujol, S., Kikinis, R., & Feng, D. (2014, April). Early diagnosis of Alzheimer’s disease with deep learning. In Biomedical Imaging (ISBI), 2014 IEEE 11th International Symposium on (pp. 1015-1018). IEEE.

[3] Pinaya, W. H., Gadelha, A., Doyle, O. M., Noto, C., Zugman, A., Cordeiro, Q., … & Sato, J. R. (2016). Using deep belief network modelling to characterize differences in brain morphometry in schizophrenia. Scientific reports, 6, 38897.

[4] van der Burgh, H. K., Schmidt, R., Westeneng, H. J., de Reus, M. A., van den Berg, L. H., & van den Heuvel, M. P. (2017). Deep learning predictions of survival based on MRI in amyotrophic lateral sclerosis. NeuroImage: Clinical, 13, 361-369.

[5] Cole, J. H., Poudel, R. P., Tsagkrasoulis, D., Caan, M. W., Steves, C., Spector, T. D., & Montana, G. (2016). Predicting brain age with deep learning from raw imaging data results in a reliable and heritable biomarker. arXiv preprint arXiv:1612.02572.

[6] Abadi, M., Agarwal, A., Barham, P., Brevdo, E., Chen, Z., Citro, C., … & Ghemawat, S. (2016). Tensorflow: Large-scale machine learning on heterogeneous distributed systems. arXiv preprint arXiv:1603.04467.

 

Secondary Project 3: Mild MTLE

Co-leads: Maria Eugenia Caligiuri & Angelo Labate (University “Magna Graecia” of Catanzaro)

Introduction:

Over the past two decades, studies focusing on non-surgical patients with mesial temporal lobe epilepsy (MTLE) have confirmed the existence of a mild form of MTLE (mMTLE). Its main features include seizure onset in adulthood, unremarkable past medical history, viscero-sensory auras, and long-term seizure freedom (>24 months) with or without antiepileptic medication [1]. Approximately one third of patients with mMTLE have MRI evidence of hippocampal sclerosis (HS), which was previously considered a hallmark of refractoriness.

Mild MTLE represents a superb resource to better delineate the biological substrates underlying the epileptic syndrome itself, and the ENIGMA-Epilepsy project gives a unique chance to carry on this investigation over a very large cohort of subjects.

Since genetic factors might have a major etiological role in mMTLE, the possibility of correlating genotype data into this proposal would be desirable.

Adult-onset mMTLE often provides few clues to suspect delayed appearance of refractoriness. Very recently, a longitudinal study has observed the long-term outcome of patients with mMTLE over a mean follow-up of twelve years [2]. Over this period, twenty-five percent of the patients, who all had mMTLE at time of enrolment, eventually developed refractoriness. At baseline, the following factors were different between patients who remained drug-responsive and those who changed disease course to refractory: age at onset, history of febrile convulsions, and the presence of HS on MRI. More in detail, mMTLE patients carrying radiological evidence of HS since recruitment displayed 3 times higher likelihood of becoming refractory later on in life than those without HS.

In light of these findings, and with the support of the ENIGMA-Epilepsy Working Group, we would like to propose a secondary analysis using existing, locally-processed T1-weighted MRI data. Initial measures of interest will be subcortical volume and cortical thickness. The phenotype to be studied will be mMTLE patients, divided on the basis of the presence/absence of HS, this being the major risk factor for development of refractoriness. Age, sex, intracranial volume (ICV) and other covariates, such as age at onset and history of febrile convulsions, will be considered.

Meta-analysis will be undertaken following the design of the first ENIGMA-Epilepsy study of subcortical volume/cortical thickness. Local analysts will be asked to re-stratify their existing patient cohorts based on the definition of mMTLE and HS, and subsequently re-run a series of linear regressions on these phenotype groups (versus healthy controls) using ENIGMA standardized R scripts. No additional image processing or quality assurance will be required.

Hypothesis:

Patients with mMTLE with HS show differential patterns of subcortical atrophy and cortical thinning compared to patients with mMTLE without HS and to healthy controls.

References:

[1] Labate A, Gambardella A, Andermann E, et al. Benign mesial temporal lobe epilepsy. Nat Rev Neurol 2011;7:237–240.

[2] Labate A, Aguglia U, Tripepi G, et al. Long-term outcome of mild mesial temporal lobe epilepsy. Neurology 2016; doi: 10.1212/WNL.0000000000002674

 

Secondary Project 4: Structural Covariance

Co-leads: Maria Eugenia Caligiuri & Angelo Labate (University “Magna Graecia” of Catanzaro)

Introduction:

In health and pathology, there are marked inter-individual differences in the structure of cortical regions. In particular, the between-subject variability is much larger if we consider the volume of a single gyrus compared to whole-brain volume [1]. The phenomenon known as ‘structural covariance’ shows that inter-individual differences in regional structure are coordinated within communities of brain regions, which fluctuate together (i.e., co-vary) in size across the population.

Networks of structural covariance have been studied in drug-resistant temporal lobe epilepsy, especially since increasing evidence supports the hypothesis that specific cortical and subcortical networks play a fundamental role in the genesis and expression of seizures [2]. However, little is known regarding possible alterations of structural co-variance patterns in patients with mild mesial temporal lobe epilepsy (mMTLE).

The main features of this phenotype include seizure onset in adulthood, unremarkable past medical history, viscero-sensory auras, and long-term seizure freedom (>24 months) with or without antiepileptic medication [3]. Approximately one third of patients with mMTLE have MRI evidence of hippocampal sclerosis (HS), which was previously considered a hallmark of refractoriness. Very recently, it has been shown that mMTLE patients with HS possess 3-times higher likelihood of becoming refractory later on in life than those without HS [4].

Mild MTLE represents a superb resource to better delineate the biological substrates underlying MTLE, in the search for correlates of the epileptic syndrome itself, and the ENIGMA-Epilepsy project gives a unique chance to carry on this investigation over a very large cohort of subjects.

Since genetic factors might have a major etiological role in mMTLE, the possibility of correlating genotype data into this proposal would be desirable.

With the support of the ENIGMA-Epilepsy Working Group, we would like to propose a secondary analysis on T1-weighted MRI data. In particular, the phenotype to be studied will be mMTLE patients, divided on the basis of the presence/absence of HS, this being the major risk factor for development of refractoriness. Cortical thickness data produced during the main ENIGMA-Epilepsy project should be used, at each site, to produce structural covariance matrices for each group (mMTLE with HS; mMTLE without HS; healthy controls) and to subsequently extract graph-based measures [5]. Novel R scripts should be developed and distributed among sites.

Age, sex, intracranial volume (ICV) and other covariates, such as age at onset and history of febrile convulsions, will be considered introducing appropriate regression models.

Hypothesis:

Patients with mMTLE stratified on the basis of HS evidence show differential patterns of subcortical atrophy and cortical thinning compared to healthy controls.

References:

[1] Alexander-Bloch et al. Imaging structural co-variance between human brain regions. Nat Rev Neuroscience 2013; doi: 10.1038/nrn3465.

[2] Bernhardt BC, Hong S, Bernasconi A and Bernasconi N (2013) Imaging structural and functional brain networks in temporal lobe epilepsy. Front. Hum. Neurosci. 7:624. doi: 10.3389/fnhum.2013.00624

[3] Labate A, Gambardella A, Andermann E, et al. Benign mesial temporal lobe epilepsy. Nat Rev Neurol 2011;7:237–240.

[4] Labate A, Aguglia U, Tripepi G, et al. Long-term outcome of mild mesial temporal lobe epilepsy. Neurology 2016; doi:10.1212/WNL.0000000000002674

[5] Bernhardt et al. Graph-theoretical analysis reveals disrupted small-world organization of cortical thickness correlation networks in temporal lobe epilepsy. Cereb cortex 2011; 21:2147-2157.

ENIGMA on social media: