Accelerated Brain Atrophy in Epilepsy: A Prospective Population‐Based Study

Study Population

We analyzed data from a prospective, community‐based longitudinal study designed to assess cerebral structural changes in a representative epilepsy population. ⁹ , ¹⁰ , ¹¹ , ¹² The sampling frame comprised all individuals registered with participating primary care practices, and the manual and electronic audit was designed to capture all cases from this community base rather than from tertiary referral pathways. Participants were recruited prospectively over a 2‐year period between June 1995 and May 1997 from a local population of 207,553 individuals under regular active surveillance by 21 general medical practices in the United Kingdom. The study included individuals aged 15 years or older with a confirmed diagnosis of established epilepsy. Case identification was optimized through an intensive manual and electronic audit of primary care records, screening for new antiseizure medication prescriptions or epilepsy‐related diagnostic codes. For subgroup analyses, the cohort was stratified by epilepsy syndrome (focal vs generalized), magnetic resonance imaging (MRI) findings (lesional vs nonlesional), and seizure status during the follow‐up period (seizure‐free vs ongoing seizures).

The control population consisted of age‐ and sex‐matched healthy controls, comprising volunteers, hospital staff, and friends or spouses of patients, recruited from the same community base.

Exclusion criteria included isolated seizures without a definitive diagnosis of epilepsy, acute symptomatic seizures only, progressive neurological disorders, significant psychiatric comorbidity, alcohol or drug abuse, MRI contraindications, MRI of poor quality (eg, movement artifacts), and failed magnetic resonance (MR) postprocessing.

All participants underwent baseline neuroimaging and were invited for a repeat scan after a target interval of 3.5 years. Patient recall was optimized through an active monitoring system involving 6‐month reviews of general practice and outpatient records. Crucially, the follow‐up scan was performed regardless of seizure outcome or clinical course, ensuring that patients without seizure remission were as likely to be rescanned as those with drug‐resistant epilepsy.

Attrition rate in the original follow‐up study was 20%. ⁹ , ¹⁰ , ¹¹ , ¹² Reasons for attrition were death (20%), infirmity (9%), inability to trace (13%), emigration (7%), pregnancy (2%), inability to tolerate scanning (6%), and refusal (43%).

Among 184 participants with 2 scans in the original follow‐up study, we excluded 21 with isolated unprovoked seizures without confirmed epilepsy, 11 with acute symptomatic seizures only, and 3 with unknown epilepsy classification, leaving 149 individuals with confirmed focal or generalized epilepsy. Imaging quality control then excluded 33 individuals: 18 because of large lesions (>30 ml by ABC/2 formula) likely to disrupt preprocessing, 1 because of a surgical resection cavity, 1 because of MRI artifact, 12 because of severe movement artifact, and 1 because of failed quality control (see below) with poor segmentation on visual review. No controls with 2 scans were excluded.

Standard Protocol Approvals, Registrations, and Patient Consents

The study protocol was approved by the Joint Ethics Committee of the National Hospital for Neurology and Neurosurgery and the Institute of Neurology, London, United Kingdom. Written informed consent was obtained from all participants prior to inclusion.

MRI Data Acquisition

Longitudinal neuroimaging data were acquired on a 1.5T General Electric Signa Horizon scanner (GE Medical Systems, Milwaukee, WI) at the Chalfont Centre for Epilepsy, Chalfont St. Peter, UK. To prevent scanner‐induced variance, baseline and follow‐up scans were performed on the same scanner without hardware or software upgrades using identical acquisition sequences. T1‐weighted structural images were acquired using a fast inversion recovery prepared 3D spoiled gradient echo (IR‐SPGR) volume sequence (TR/TE/TI/NEX = 17.4/4.2/450/1; flip angle = 20 degrees; matrix = 256 × 192; and field of view [FOV] = 24 × 18 cm). The acquisition consisted of 124 contiguous coronal 1.5‐mm‐thick slices, providing whole‐brain coverage for detailed morphometric analysis.

Image Processing and Longitudinal Morphometry

We estimated total intracranial volume, grey and white matter volumes, cortical thickness, and regional subcortical volumes for all scans in all cohorts using the fully automated, ¹³ validated, ¹³ , ¹⁴ , ¹⁵ and reliable ¹³ , ¹⁵ Computational Anatomy Toolbox (CAT12; Jena University Hospital, Germany; https://neuro-jena.github.io/cat/) within Statistical Parametric Mapping 12 (SPM12; Wellcome Centre for Human Neuroimaging, United Kingdom; https://www.fil.ion.ucl.ac.uk/spm/software/spm12/) running in MATLAB R2023b (The MathWorks, Natick, MA) with an inverse‐consistent longitudinal registration approach. ¹⁶ The CAT12 toolbox previously showed excellent test–retest reliability (R ² = 0.986) and was validated against other cortical surface reconstruction methods, showing fewer thickness measurement errors than comparable approaches. ¹⁴ , ¹⁵

Cortical thickness was computed using CAT12's projection‐based thickness method using standard settings, which derives white matter distance from tissue segmentation obtained using geodesic shooting and projects local maxima of thickness to grey matter voxels, effectively accounting for partial volume effects and sulcal geometry. The pipeline comprises topology correction using spherical harmonics, spherical mapping to a common coordinate system, diffeomorphic surface registration for inter‐subject alignment, and subsequent smoothing with a 12 mm surface‐based kernel.

Subcortical regional volumes in 5 regions (hippocampus, amygdala, thalamus, pallidum, and putamen) were calculated by projecting grey matter volume (GMV) segmentations onto the Hammers brain atlas in template space. ¹⁷

All data underwent CAT12‐based quality control, and scans with signs of excessive misalignment, misregistration, inaccurate estimations, or large lesions that may disrupt processing were excluded as detailed above. Image quality was assessed by evaluating image noise, inhomogeneities, and resolution within the CAT12 quality assurance framework.

Statistical Analysis

We estimated the rate of GMV, cortical thickness, and subcortical volume change using linear mixed‐effects models (Fig 1), a flexible framework for longitudinal analysis of multiple repeated measurements per subject with irregular measurement intervals. Consistent with neuroimaging convention, we operationalized reductions in GMV and cortical thickness as “atrophy,” acknowledging that in the absence of histological validation, these macroscopic volume losses may reflect reduction of neuropil rather than neuronal loss, per se. ¹⁸ Vulnerable cortical regions were defined as those areas that showed significantly accelerated cortical thinning in epilepsy compared with healthy controls (Fig 2).

For each non‐overlapping comparison, we fitted a separate model with the same architecture. The primary term of interest was the interaction between the variable of interest (ie, group allocation or clinical variable) and age at MRI scan; the corresponding p values therefore test whether longitudinal slopes differed between groups or across clinical variables. All models included a random effect of participant, fixed effects of age at scan, sex, and group, and the relevant interaction with age at scan. With this approach, we were able to test for within‐participant morphometric changes over time while correcting for baseline demographic differences and for different interscan intervals. Volumetric data (GMV and subcortical volumes) were additionally corrected by including total intracranial volume as a covariate. Models comparing clinical variables within participants with epilepsy were additionally corrected for seizure frequency in between MRI scans, history of tonic–clonic seizures, presence of hippocampal sclerosis, duration of epilepsy at baseline, and number of antiseizure medications at baseline, except when the covariate itself was the variable of interest.

We report findings considered significant at p < 0.05 corrected for multiple comparisons using random field theory for nonisotropic images on a cluster level ¹⁹ for cortical thickness and after Bonferroni correction for the 10 subcortical volumes tested (corrected p‐values shown in manuscript).

We performed statistical analyses using R software version 4.4.1 (R Foundation for Statistical Computing, Vienna, Austria). Cortical thickness and subcortical volume calculations were performed using SurfStat (https://www.math.mcgill.ca/keith/surfstat/) in MATLAB.

Study Population

The final analysis comprised 116 individuals with confirmed epilepsy (median age at baseline scan = 34 years, interquartile range = 25–49; 63 female subjects) and 89 healthy controls (median age = 35 years, interquartile range = 27–48; 40 female subjects; see the Table 1). Within the epilepsy cohort, 85 individuals (73%) were classified as having focal epilepsy, whereas 31 (27%) had generalized epilepsy. Structural MRI abnormalities, including hippocampal sclerosis, were identified in 40 participants (34%). During the follow‐up interval, 90 individuals (78%) continued to experience seizures, whereas 26 (22%) remained seizure‐free. Among individuals with focal epilepsy, seizure focus laterality was left‐sided in 16, right‐sided in 8, bilateral in 13, and unknown or undetermined in 48.

Global and Regional Rates of Atrophy

Analysis of global GMV reduction demonstrated a significantly faster rate of loss in people with epilepsy (2.7 ± 0.2 ml/year; p = 0.01 for the epilepsy‐by‐age‐at‐scan interaction) compared with healthy controls (2.4 ± 0.2 ml/year; see Fig 1A); rate difference = 0.27 ml/year (95% confidence interval = 0.06–0.49). Subgroup stratification indicated that this accelerated GMV loss compared with healthy aging was driven by specific phenotypes: it was significant in focal epilepsy (2.7 ± 0.2 ml/year; p = 0.005) but not generalized epilepsy (2.5 ± 0.2 ml/year; p = 0.37); in those with ongoing seizures (2.7 ± 0.2 ml/year; p = 0.01) but not seizure‐free individuals (2.6 ± 0.2 ml/year; p = 0.23); and in MRI‐positive (2.8 ± 0.2 ml/year; p = 0.004) but not MRI‐negative individuals (2.6 ± 0.2 ml/year; p = 0.08). There were no differences in atrophy rates between male and female patients with epilepsy.

Given the heterogeneous distribution of epilepsy‐related structural changes, we further quantified atrophy rates within “vulnerable cortical regions,” areas defined by the vertex‐wise analysis as showing significantly accelerated thinning in the overall epilepsy group (see Fig 2); this analysis was used to summarize the magnitude of change in affected regions rather than as an independent topographic test. In these regions, the rate of cortical thinning was markedly greater in people with epilepsy (5.3 ± 0.6 μm/year; p < 0.001) relative to healthy controls (3.4 ± 0.5 μm/year; see Fig 1B); rate difference = 1.9 μm/year (95% confidence interval = 1.2–2.5). Unlike global GMV, accelerated thinning in vulnerable regions was more pervasive across subtypes: it was significant in both focal (5.3 ± 0.6 μm/year; p < 0.001) and generalized epilepsies (4.7 ± 0.7 μm/year; p = 0.005), and in both MRI‐positive (5.7 ± 0.7 μm/year; p < 0.001) and MRI‐negative cases (5.0 ± 0.7 μm/year; p < 0.001). Among those with ongoing seizures, thinning was accelerated (5.4 ± 0.6 μm/year; p < 0.001), whereas seizure‐free individuals showed a non‐significant trend toward acceleration (4.5 ± 0.8 μm/year; p = 0.07). Among the 26 seizure‐free individuals, 22 had focal epilepsy and 4 had generalized epilepsy.

Topography of Progressive Damage

Vertex‐wise longitudinal morphometry revealed specific patterns of accelerated atrophy in people with epilepsy compared with healthy controls (see Fig 2). Significant clusters were identified in the right (1,021 vertices; p < 0.001) and left (605 vertices; p = 0.02) parieto‐occipital junctions, as well as the right (557 vertices; p = 0.004) and left (509 vertices; p = 0.008) medial pericentral cortices. Subcortical analysis demonstrated significant progressive volume loss in the left (t = 3.7; p < 0.001) and right (t = 2.2; p = 0.01) thalami, and the left (t = 3.0; p = 0.001) and right (t = 2.7; p = 0.003) hippocampi. No areas of progressive hypertrophy ²⁰ were detected in the epilepsy group compared with healthy aging in controls.

When stratified by seizure status, individuals with ongoing seizures (n = 90) exhibited a widespread topographic pattern similar to the overall cohort. In contrast, seizure‐free individuals (n = 26) displayed a restricted phenotype, with accelerated atrophy confined to the right hippocampus (t = 2.1; p = 0.02). The laterality of epilepsy also influenced the spatial distribution of damage (Fig 3). People with left‐lateralized epilepsy showed accelerated atrophy restricted to the left hemisphere, whereas those with right‐lateralized epilepsy exhibited accelerated atrophy of the right hippocampus and bilateral thalami, with a right‐sided predominance.

Regarding epilepsy syndrome (Fig 4), no significant clusters of accelerated atrophy were detected in the generalized epilepsy group (n = 31). Conversely, people with focal epilepsy (n = 85) displayed the characteristic pattern observed in the overall group, with additional involvement of the superior frontal gyrus. Within focal epilepsy, MRI‐positive individuals (n = 36) showed extensive changes similar to the overall focal cohort, whereas MRI‐negative focal epilepsy (n = 49) was associated with a more limited pattern involving only the right medial pericentral cortex and the left thalamus.

Clinical Determinants of Progressive Atrophy

To delineate the clinical drivers of progressive damage, we performed multivariable linear mixed effects analyses within the cohort of people with epilepsy (n = 116). Higher seizure frequency was associated with accelerated atrophy of the bilateral putamina and a nonsignificant trend toward faster atrophy of the right frontal operculum; conversely, it was associated with slower atrophy of the right hippocampus (Fig 5A). A history of tonic–clonic seizures was predictive of accelerated atrophy in the right hippocampus, thalamus, and amygdala (Fig 5B). The presence of hippocampal sclerosis was associated with more rapid atrophy of the hippocampi and the left amygdala (Fig 5C). Shorter duration of epilepsy was associated with accelerated atrophy of the right putamen and bilateral pallida (Fig 5D), whereas a higher baseline load of antiseizure medications was associated with accelerated atrophy of the right hippocampus and thalamus (Fig 5E).

Direct comparisons between clinical subgroups further clarified these associations (Fig 6). People with focal epilepsy exhibited significantly accelerated atrophy in the bilateral thalami and left hippocampus compared with those with generalized epilepsy (see Fig 6A). Individuals with ongoing seizures showed more rapid atrophy of the bilateral putamina but slower atrophy of the right hippocampus compared with seizure‐free individuals (see Fig 6B). Finally, MRI‐positive individuals demonstrated significantly accelerated atrophy in the bilateral hippocampi, thalami, and amygdalae, along with a nonsignificant trend in the left precentral gyrus, compared with MRI‐negative cases (see Fig 6C).