Exploring Cortical and Hippocampal Changes in Temporal Lobe Epilepsy Using Automated MRI Segmentation Techniques

Introduction

In 2010, the International League Against Epilepsy (ILAE) redefined epilepsy, emphasizing its network attributes as a neurological disorder marked by abnormal discharges originating in local brain networks and dynamically spreading to other areas. Temporal lobe epilepsy (TLE) is a prevalent form of drug-resistant epilepsy, mainly localized in the temporal lobe and hippocampal structures, with hippocampal sclerosis (HS) being a leading pathological cause.¹ Surgical resection proves effective, alleviating symptoms in 70% to 90% of patients.² However, TLE’s structural damage extends beyond the temporal lobe and hippocampus, impacting various regions like the thalamus, insula, basal ganglia, frontal lobe, posterior cingulate, cerebellum, and fusiform gyrus.³ This broader involvement correlates with cognitive impairments beyond temporal lobe functions, including declines in executive function, cognitive processing speed, attention, and diffuse reductions in neocortical metabolism and functional connectivity.⁴ Understanding diverse brain region damage in TLE patients is crucial for comprehending the disease’s mechanisms, guiding clinical intervention, and prevention.

Advanced magnetic resonance imaging (MRI) technology and related research have introduced an array of MRI techniques for epilepsy diagnosis and treatment, especially in diagnosing HS. Techniques like multiplexed sensitivity encoding diffusion weighted imaging (MUSE-DWI) and three-dimensional pseudo-continuous arterial spin labeling (3D-pCASL) significantly enhance preoperative detection rates of HS by assessing changes in blood flow.⁵ However, about 20–30% of epileptic foci remain undetected in conventional 3.0T MRI scans, known as nonlesional temporal lobe epilepsy (TLE-NL).⁶ Studies⁷ suggest varying degrees of brain microstructural damage in TLE-NL patients, emphasizing the need for further evaluation of these undetected microstructural injuries.

Cortical thickness serves as a reliable indicator of brain structural changes. While manual delineation of cortical thickness is time-consuming and prone to low repeatability, automated segmentation software, such as FreeSurfer, minimizes errors caused by subjective factors.⁸ Proven accurate and reliable for measuring cortical thickness and volume, FreeSurfer is particularly sensitive to hippocampal atrophy in TLE compared to other methods.⁹ The software segments the brain hemisphere into 31 regions, subdividing frontal, parietal, temporal, and occipital lobes into specific gyri and sulci (Figure 1). The latest guidelines propose recommendations for MRI usage and outline the Harmonized Neuroimaging of Epilepsy Structural Sequences (HARNESS) scheme. The 3D-T1WI-MPRAGE sequence, a commonly used sequence,¹⁰ aligns with FreeSurfer’s automatic segmentation capabilities.

Figure 1 (A) The names and corresponding color codes of 31 cerebral hemispheres in the left hemisphere in FreeSurfer. (B) The name and corresponding color code of the hippocampus subregion.

In summary, this study aims to utilize FreeSurfer software based on the 3D-T1WI-MPRAGE sequence to automatically segment and measure cortical thickness or volume in various brain regions of HS and TLE-NL patients. The investigation seeks to explore the application value of MRI automatic segmentation techniques in assessing the extent of damage in different regions for HS and TLE-NL patients.

Materials and Methods

Study Population

This study, aligned with the Helsinki Declaration, received approval from the Ethics Review Committee of the General Hospital of Ningxia Medical University (Ethics Approval No: KYLL-2021-0295). Each participant was informed about the study content, consent was obtained, and a written informed consent was signed. Between January 2021 and September 2023, 48 patients suspected of epilepsy underwent integrated epilepsy scans at our hospital, with 48 diagnosed with TLE based on clinical presentations and EEG, following ILAE 2017 diagnostic criteria. Two neuroradiologists jointly diagnosed the patients using the 3.0T MRI HARNESS scan protocol.

30 HS patients and 18 TLE-NL patients were included. Additionally, 30 healthy volunteers were randomly recruited from the normal population as a control group.

Inclusion criteria for the patient group were: (1) symptomatic and EEG-confirmed left TLE; (2) MRI showing left HS or normal findings; (3) age between 18 and 60 years; (4) MRI conducted during the interictal period. Exclusion criteria included: (1) neurological or psychiatric disorders or family history; (2) other conditions causing epileptic symptoms; (3) congenital brain malformations; (4) poor MRI image quality, inability for automated segmentation, or mismatched segmentation.

Healthy control participants were recruited from the local community and from hospital staff. Eligible participants were aged 18–59 years, right-handed, and had no history of neurological or psychiatric disorders, including epilepsy, stroke, or dementia. Controls were also required to have no significant cognitive impairments, as assessed by a standard neuropsychological screening battery, and no MRI abnormalities. Individuals with a history of substance abuse, major medical conditions (eg, cardiovascular disease), or any contraindications for MRI scanning were excluded from participation.

MRI Scanning Protocol

A GE SIGNA Architect 3.0T MRI scanner (Signa Architect, GE Healthcare) with a 48-channel phased-array head coil was used. All participants underwent our hospital’s integrated epilepsy MRI scanning protocol, with the main sequence parameters as follows: (1) Axial 3D T1-weighted magnetization-prepared rapid acquisition gradient echo sequence (3D T1WI-MPRAGE), isotropic voxels (1.0 mm×1.0 mm×1.0 mm), FOV 256 mm×256 mm, TR 7.7 ms, TE 3.1 ms, flip angle 8°, one excitation, bandwidth 1.25 hz; (2) Coronal T2-weighted imaging perpendicular to the long axis of the hippocampus, TR 2601 ms, TE 85 ms, slice thickness 2.0 mm, interslice gap 1.0 mm, flip angle 111°, four excitations, bandwidth 50 hz.

Image Analysis

Two experienced neuroradiologists (J.L and L.W, with over 7 years of experience each) independently analyzed all images in a blinded manner. Using high-resolution coronal T2-weighted images, they visually assessed hippocampal morphology, signal intensity, and internal structure. The MRI-HS diagnostic criteria included: (1) direct signs of reduced overall hippocampal volume, increased signal intensity on T2-weighted or T2 fluid attenuated inversion recovery (T2-FLAIR); (2) indirect signs of internal structure loss, blurred or absent stratum radiatum, flattened cornu ammonis, enlargement of the temporal horn of the lateral ventricle, ipsilateral hippocampal atrophy, and temporal lobe atrophy. The diagnosis for MRI-negative was based on visual analysis confirming normal brain structures. In case of disagreement, a senior radiologist (B.C) reassessed, and consensus was reached through discussion.

Image Post-Processing

In this study, FreeSurfer (http://surfer.nmr.mghharvard. edu. was employed for the segmentation of the cerebral cortex and hippocampal subfields. The segmentation process is outlined as follows: The 3D T1WI images were first converted from the DICOM format to.nii/.nii.gz format using the MRI cron software. Subsequently, these images were processed in the Linux system using FreeSurfer software version 7.3.2 for whole-brain segmentation (recon-all). This comprehensive segmentation includes 31 steps, encompassing head motion correction, non-uniform intensity normalization, and other procedures, all executed through the recon-all script for automation [http://ftp.nmr.mgh.harvard.edu/fswiki/recon-all]. Building upon recon-all, the segmentHA_T1.sh script was applied for the segmentation of hippocampal subfields,¹¹ distinguishing the head, body, and tail of the hippocampus. The thickness of each gyrus and the volume of each hippocampal subfield were obtained in the stats output file.

Statistical Analysis

Statistical analysis was conducted using SPSS 26.0 software. Before conducting statistical tests, we assessed whether statistics including the differences in hippocampal subfield volumes and cortical thickness between patients (TLE subtypes and controls) were normally distributed. We tested the normality of these differences using both graphical methods (Q-Q plots) and formal statistical tests (Shapiro–Wilk test). Descriptive statistics, mean ± standard deviation for normally distributed data, and median (Q1, Q3) for non-normally distributed data, were employed. Independent sample t-tests (for normally distributed data) or independent sample Wilcoxon tests (for non-normally distributed data) were performed to compare the differences in cortical thickness of various brain regions on the left side, hippocampal total volume, and volumes of hippocampal subfields (head, body, tail) between the control group and the TLE-NL group, control group and HS group, and between the TLE-NL and HS groups. A significance level of P < 0.05 was considered statistically significant. Post-hoc power analysis was conducted to assess the statistical power of the study. Parameters such as the observed effect size, sample size, and significance level were used to determine that the study achieved a power of ≥ 0.80, ensuring sufficient sensitivity to detect differences in cortical thickness and hippocampal subfield volumes between groups.

Results

Characteristics of the Study Participants

A total of 48 TLE patients were included in this study, including 30 hS patients (16 male, 14 female; average age, 32.30±11.38 years; range, 18 ~ 55 years); 18 TLE patients (10 male, 8 female; average age, 33.55±14.25 years; range, 18 ~ 59years). Thirty healthy volunteers (17 male, 13 female; average age, 28.03±10.11 years; range, 18 ~ 58 years) were recruited. No statistically significant differences in age or gender were observed among the three groups (age: Z¹ = −1.095, P¹ = 0.273; Z² = −1.487, P² = 0.137; Z³ = −0.032, P³ =0.974; gender: χ² = 2.468, P = 0.291). The demographic characteristics, electrophysiological findings, and semiology of all participants are shown in Table 1.

Table 1 The Demographics, Electrophysiological Findings, and Semiology of the Study Participants

Differences in Cortical Thickness and Hippocampal Subfield Volumes in the HC

In the control group, normal distribution characterizes hippocampal volumes on both sides. Paired-sample t-tests unveil significant differences in total volume, head, and body of the hippocampus between the left and right sides. Regarding cortical thickness measurements, some conform to a normal distribution on both sides, while others do not. Employing paired-sample t-tests and paired-sample Wilcoxon rank-sum tests, we identify 11 cortical thickness measurements with significant differences between the left and right sides. These include the posterior cingulate cortex, olfactory cortex, cingulate gyrus, inferior parietal lobule, lateral occipital gyrus, middle temporal gyrus, precuneus, posterior cingulate gyrus, superior frontal gyrus, middle frontal gyrus, and inferior frontal gyrus. Significance is set at P < 0.05, and detailed results are provided in Table 2 and 3.

Table 2 Results in Total Volume and Subregion Volume Values of the Left and Right Hippocampus in the Control Group

Table 3 The Thickness of 31 Cerebral Cortexes on the Left and Right Sides of the Control Group

Differences in Cortical Thickness Among HC, TLE-NL, and HS Groups in the Left Hemisphere

Cortical thickness across various regions of the left hemisphere deviated from a normal distribution in all three groups. Employing independent sample Wilcoxon rank-sum tests for statistical analysis, we identified significant differences in six cortical thickness regions between the control and TLE-NL groups. These regions included the medial frontal gyrus (head and tail), superior frontal gyrus, triangular part of the inferior frontal gyrus, paracentral lobule, and posterior cingulate gyrus. Notably, the frontal lobe encompassed the medial frontal gyrus (head and tail), superior frontal gyrus, and triangular part of the inferior frontal gyrus, while the posterior cingulate gyrus belonged to the parietal lobe. Fifteen cortical thickness regions exhibited significant differences between the control and HS groups, spanning areas such as the medial frontal gyrus (head and tail), lateral orbital gyrus, triangular part of the inferior frontal gyrus, lateral orbital gyrus (frontal lobe), olfactory cortex (temporal lobe), fusiform gyrus, superior temporal gyrus, middle temporal gyrus, inferior temporal gyrus, parahippocampal gyrus (temporal lobe), posterior cingulate gyrus, and marginal gyrus (parietal lobe). Additionally, differences in cortical thickness between the TLE-NL and HS groups were observed in the olfactory cortex (temporal lobe) and the tail of the rostral cingulate gyrus (frontal lobe). Detailed information can be found in Table 4, Figures 2–4, with statistical significance set at P < 0.05.

Table 4 Cortical Thickness of 31 Left Brain Regions in the Three Groups and Comparison Between Groups

Figure 2 6 different thinning cortical color codes and names of TLE-NL patients. Abbreviation: TLE-NL, nonlesional temporal lobe epilepsy.

Figure 3 15 thinning cortical color codes and names in HS patients. Abbreviation: HS, hippocampal sclerosis.

Figure 4 (A and B) Comparison of cortical thickness in various regions between HC, TLE-NL and HS groups. Abbreviations: HC, healthy control; TLE-NL, nonlesional temporal lobe epilepsy; HS, hippocampal sclerosis.

Differences in Hippocampal Total Volume and Subfield Volumes among Control, TLE-NL, and HS Groups in the Left Hemisphere

Volumes of the hippocampus total, head, body, and tail deviated from a normal distribution in all three groups. Employing independent sample Wilcoxon rank-sum tests, no statistically significant differences emerged in hippocampal total volume, head, body, and tail volumes between the control and TLE-NL groups. However, notable differences were observed between the control and HS groups, as well as between the TLE-NL and HS groups, across all hippocampal measurements (total volume, head, body, and tail volumes), with P < 0.01. Specific values can be found in Table 5 and Figure 5.

Table 5 Total Volume and Subregion Volume of Left Hippocampus in Three Groups and Comparison Between Groups

Figure 5 Comparison of total hippocampal volume and volume of each part in HC, TLE-NL and HS groups. Abbreviations: HC, healthy control; TLE-NL, nonlesional temporal lobe epilepsy; HS, hippocampal sclerosis.

Discussion

This study, encompassing healthy controls and two TLE subtypes (hippocampal sclerosis - HS and MRI-negative cases), delves into distinctive gray matter abnormalities indicated by structural MRI studies. Notably, patients with medial temporal lobe sclerosis exhibit pronounced abnormalities in the ipsilateral medial temporal lobe, along with frontal, parietal lobes, basal ganglia, and thalamus involvement. Conversely, MRI-NL patients manifest anomalies in the ipsilateral inferior temporal lobe cortex, extending to the insula and frontal lobe regions.^12–15

For the diagnosis of hippocampal sclerosis, this study adopted the ILAE 2017 diagnostic criteria. Regarding the detection of hippocampal sclerosis from MRI images, we used blind analysis by radiologists and visual assessment of the hippocampus, as in previous studies.^5,16 Princich et al¹⁶ reported that measuring the total hippocampal volume using automatic classifiers such as FreeSurfer can improve the detection rate of hippocampal sclerosis (AUC: >0.960). This study used FreeSufer to detect changes in hippocampal subregion volume in hippocampal sclerosis.

Asymmetries in the Cerebral Cortex of Healthy Humans

Consistent with prior research,^17,18 our study underscores cortical thickness variations across diverse brain regions in the healthy control group (Table 3). Additionally, it elucidates differences in hippocampal total volume and subfield volumes (Table 2), aligning with recognized asymmetries in the human cerebral cortex. Intriguingly, asymmetry in cerebellar structures is discerned in HS patients, particularly those with left TLE-HS.¹⁹ Unilateral hippocampal sclerosis affects not only the cerebral cortex on the ipsilateral side but also the cerebral cortex on the contralateral side. Moreover, there is asymmetry in the thickness of the cerebral cortex on both sides. If left and right hippocampal sclerosis are included in the study at the same time, the problem will be complicated. It is difficult to determine whether it is asymmetry or changes in cortical thickness caused by hippocampal sclerosis. To precisely delineate unilateral HS’s impact, we exclusively enrolled left-sided HS patients to improve consistency, concentrating on the extent of involvement in the left hemisphere.

Changes in Hippocampal Volume and Cortical Thickness in TLE-NL Patients

Approximately 30% of TLE cases fall under the category of TLE-NL,³ with existing research pointing to structural cortical alterations in these patients. Traditional visual analysis and routine MRI might not unveil apparent changes, making FreeSurfer software instrumental for precise structural quantification. This study underscores that TLE-NL patients exhibit cortical thinning in specific regions, primarily affecting the ipsilateral frontal lobe cortex. Notably, thinning involves the medial frontal gyrus (head and tail), superior frontal gyrus, triangular part of the inferior frontal gyrus (frontal lobe), paracentral lobule (frontal and parietal lobes), and posterior cingulate gyrus (parietal lobe). Despite the absence of significant hippocampal volume changes, cortical thinning in TLE-NL predominantly impacts the frontal lobe, with limited parietal lobe involvement, while the temporal lobe remains unaffected—a trend in line with previous findings.¹¹

TLE emerges as a complex interplay among multiple brain regions, influenced by the propagation pathways of epileptic discharges. The well-established phenomena of discharges spreading through neural networks to other brain areas and evolving into generalized seizures add complexity to understanding TLE.²⁰ Presently, conclusive evidence regarding the sequential relationship between temporal lobe cortex changes and those extending beyond in TLE patients is lacking, necessitating further investigation for comprehensive insights.

Changes in Hippocampal Volume and Cortical Thickness in TLE-HS Patients

Approximately 65% of TLE cases manifest as TLE-HS.¹⁵ While TLE was historically perceived as a hippocampal disorder, it’s now acknowledged as a network disease with implications for various cortical and subcortical structures beyond the temporal lobe.²¹ This study delves into cortical thickness alterations across diverse brain regions in TLE-HS patients, providing a comprehensive examination of structural changes. Findings not only unveil reduced hippocampal volume but also cortical thinning in areas like the medial frontal gyrus (head and tail), superior frontal gyrus, triangular part of the inferior frontal gyrus (frontal lobe), lateral orbitofrontal cortex, orbital part of the inferior frontal gyrus, head and tail of the anterior cingulate gyrus (frontal lobe), olfactory cortex, insula, superior temporal gyrus, middle temporal gyrus, inferior temporal gyrus, parahippocampal gyrus (temporal lobe), posterior cingulate gyrus, precuneus (parietal lobe), and superior parietal gyrus (parietal lobe).¹² The identified regions of cortical thinning align with the commonly observed propagation pattern during intracranial depth electrode monitoring in focal temporal lobe epilepsy patients, involving sequential spread to the ipsilateral frontal lobe, contralateral frontal lobe, and contralateral temporal lobe.²²

The frontal lobe cortex, crucial for decision-making and flexible behavior,²³ and the temporal lobe cortex, involved in high-level cognitive functions,^24,25 both undergo thinning in TLE. This dual impact affects decision-making abilities and cognitive functions, significantly impacting patients’ quality of life. Concurrent damage to the brain cortex and white matter fiber bundles²⁶ collectively shapes brain function, suggesting that future research could explore the combined impact for a more nuanced understanding of TLE development and mechanisms. Such insights could serve as imaging-based evidence for refining clinical treatment strategies.

Although we observed some cortical thickness changes in TLE-NL and HS patients, we did not observe scalp EEG differences in corresponding regions. We believe that the corresponding changes in cortical thickness may precede EEG changes, which requires further study.

In this study, an 18-year-old male patient with left hippocampal sclerosis, whose hippocampal volume was reduced mainly in the head of the hippocampus, completely disappeared after neurosurgery surgically removed the head of the hippocampus. Regarding the correlation between the treatment effect and the volume changes in the head, body, and tail subregions of the hippocampus, there is only one case in this study, and more cases are needed for further verification. The identification of distinct cortical and hippocampal changes in different TLE subtypes using automated MRI-based segmentation techniques offers a non-invasive method for better understanding the pathophysiology of TLE. This could lead to more accurate diagnostic tools, particularly for TLE-NL, which is often difficult to diagnose using conventional imaging. Furthermore, these findings could inform personalized treatment strategies, targeting specific regions of the brain that are affected in each subtype.

While our study provides valuable insights into the structural changes in TLE, there are limitations such as the retrospective nature of the analysis and the relatively small sample size. Future studies with larger cohorts, longitudinal designs, and the incorporation of functional imaging techniques are needed to validate these findings and explore their potential role in predicting treatment outcomes.

Conclusion

To conclude, TLE manifests cortical thinning, primarily in the frontal and temporal lobes, with marginal involvement in the parietal lobe. TLE-NL exhibit frontal lobe cortex alterations, sparing the parietal lobe, and lack significant hippocampal volume changes. In contrast, patients with HS display more extensive cortical changes, impacting the frontal, temporal, and a limited portion of the parietal lobes, accompanied by a more pronounced reduction in hippocampal volume. These findings equip clinicians with nuanced imaging-based insights into microstructural brain alterations, facilitating a comprehensive preoperative assessment. The segmentation precision of FreeSurfer software emerges as instrumental in detecting subtle changes in brain cortical structures.