Abstract
Purpose
Precise lesion localisation is central to favourable surgical outcomes in patients with drug-resistant epilepsy due to focal cortical dysplasia. We validated the Anatomy-Corrected Asymmetry Index of fluorodeoxyglucose positron emission tomography to objectively delineate regions of hypometabolism and predict surgical outcome in patients with focal cortical dysplasia.
Methods
We first established normative Anatomy-Corrected Asymmetry Index thresholds using five independent and heterogeneous healthy control cohorts (n = 179) to assess reproducibility. We subsequently validated our findings in 40 pathologically confirmed focal cortical dysplasia patients who underwent resective surgery. The primary analysis assessed the association between the extent of resection of defined hypometabolic clusters and 24-month seizure freedom (Engel Class I).
Results
After excluding subjects with segmentation artefacts to ensure accurate threshold definition, we established a normative threshold of -19 with a minimum cluster size of 2.0 mL. In the patient cohort, 29 subjects had high-quality segmentations. Patients with at least one localising cluster achieved significantly higher seizure freedom rates compared to those without clusters (76.5% vs. 22.2%, P = 0.014). The percentage of resected hypometabolic volume was significantly associated with Engel Class I outcome (P = 0.011). Sensitivity analysis demonstrated that a more permissive threshold of -16 identified localising clusters in two additional seizure-free patients who were false negatives at the standard threshold.
Conclusion
The Anatomy-Corrected Asymmetry Index provides a standardised metric that predicts seizure freedom in focal cortical dysplasia epilepsy surgery. Clinical implementation requires rigorous quality control of anatomical segmentations to ensure diagnostic accuracy.
Supplementary Information
The online version contains supplementary material available at 10.1007/s00259-026-07864-9.
Keywords: Epilepsy surgery, Refractory epilepsy, Neuroimaging, Prognosis, Image quantification
Introduction
Focal cortical dysplasia (FCD) is a leading cause of drug-resistant epilepsy amenable to surgery, where precise lesion localisation is central to a successful outcome [1–4]. The presurgical evaluation relies on the concordance of multimodal data, including EEG, structural MRI, and functional imaging such as 18F-fluorodeoxyglucose positron emission tomography (FDG-PET) [3]. For structural MRI, multicentric efforts like the MELD project are developing harmonized quantitative frameworks to improve the detection of subtle FCD lesions [5]. Similarly, other quantitative approaches such as MAP18 have demonstrated high accuracy for delineating FCD lesions [6, 7]. However, a comparable, validated framework for FDG-PET is currently lacking, although recent efforts to harmonize databases are promising [8].
Clinical PET interpretation often relies on qualitative visual assessment by a nuclear medicine expert, and existing methods are heterogeneous, frequently depending on internal control databases that can introduce significant inter-site variance [8]. Asymmetry indices offer inherent advantages in robustness across datasets compared to standardised uptake value ratio (SUVR) methods, but anatomical variations can lead to important artefacts [9–11]. To address these issues, we aimed to validate the Anatomy-Corrected Asymmetry Index (ACAI), an in-house developed method, in FDG-PET. ACAI provides inherent advantages compared to other asymmetry indices by correcting for normal grey matter structural asymmetries [11, 12].
We first evaluated the ACAI technique on heterogeneous control datasets with different patient age, scanner acquisition protocols and MRI acquisition protocols to assess its robustness and evaluate any technique-related biases. Our aim was to develop a standardised FDG-PET ACAI methodology to assess and delineate hypometabolism in epilepsy with a broad applicability. Secondly, we tested its predictive value for surgical outcomes in an FCD cohort.
Materials and methods
Subjects
For the development and validation of ACAI thresholds, we assembled five independent adult healthy control cohorts (total n = 179). These comprised two in-house cohorts (one younger in-house cohort [Y_IHC] and one older in-house cohort [O_IHC]) scanned at our institution and three external cohorts from publicly available databases (ADNI, CERMEP_IDB, and the Marseille cohort), ensuring heterogeneity in age, scanners, and acquisition protocols [8, 13–16].
We retrospectively included 40 consecutive adult (> 18 years old) patients with refractory epilepsy who underwent resective surgery at University Hospital Leuven between January 2002 and January 2022. All had pathology-confirmed FCD. Each patient received a comprehensive pre-surgical evaluation, including at least one week of EEG-monitoring, structural MRI, ictal and interictal single-photon emission computed tomography if available, interictal FDG-PET (defined as the absence of recent clinical seizures prior to injection and no electrographic seizures on EEG up to 30 min after FDG injection), semiological evaluation of seizures, neuropsychological assessment, and invasive EEG recording in selected cases. A multidisciplinary meeting involving a nuclear medicine specialist, neurosurgeon, neurologist, psychiatrist, and psychologist discussed all patients prior to surgery. Variables collected included patient demographics (sex, age at seizure onset, age at surgical resection), pre-surgical clinical assessment (anti-seizure medications (ASM), lobar location, lateralisation), MRI findings (one congruent, resected lesion; normal MRI; multiple or non-congruent/non-resected lesions), PET findings (qualitative FDG-PET hypometabolism), surgical outcomes (Engel classification at 24 months, ASM at follow-up), and pathological diagnosis (FCD I, IIa or IIb) [17–19]. We determined FCD subtypes based on pathological reports; where the subtype was not explicitly stated, we defined FCD Type IIa by the presence of dysmorphic neurons and Type IIb by the additional presence of balloon cells [18, 19]. MRI and FDG-PET findings were assessed qualitatively; we considered the scan protocols established before the multidisciplinary evaluation. For outcome analysis, 24-month Engel classification was dichotomized into good outcome (Engel Class I) versus poor outcome (Engel Class > I).
The patient cohort included patients with pathology-confirmed FCD I and II, as the histological subtype is determined postoperatively.
The study protocol was reviewed and approved by the Ethics Committee UZ/KU Leuven. All participants in the in-house healthy control cohorts provided written informed consent. For the retrospective patient cohort, the Ethics Committee approved the use of data based on the institutional opt-out consent procedure.
Image acquisition
T1-weighted MRI scans were acquired for all in-house (3 T), ADNI (3 T), CERMEP_IDB (1.5 T), and Marseille (1.5 T) datasets. In-house controls were scanned on GE SIGNA PET/MR systems (TR ≈ 8.5 ms, TE ≈ 3.2 ms, TI = 450 ms, flip angle = 12°, voxel size 1–1.2 mm3). ADNI datasets included multiple 3D T1-weighted sequences: IR-SPGR and accelerated IR-FSPGR (GE; TR ≈ 7–7.7 ms, TE ≈ 2.8–3.1 ms, TI = 400 ms, flip angle = 11°, voxel size ≈ 1–1.2 mm3), and MPRAGE variants (Siemens/Philips; TR ≈ 6.8–2300 ms, TE ≈ 3 ms, TI = 0–900 ms, flip angle = 9°, voxel size ≈ 1 mm3). CERMEP_IDB subjects were acquired on a Siemens Sonata 1.5 T scanner using 3D MPRAGE (TR = 2400 ms, TE = 3.55 ms, TI = 1000 ms, flip angle = 8°, voxel size = 1.2 mm3). Marseille subjects were acquired on a Siemens SymphonyTim 1.5 T scanner (TR = 1880 ms, TE = 2.92 ms, TI = 1100 ms, flip angle 15°, voxel size = 1 mm3).
PET data for in-house controls were acquired in list mode, corrected for decay, scatter, and attenuation (MR-derived zero-time echo [ZTE] maps), and reconstructed using VPFX-S (OSEM with TOF and resolution modelling; four iterations, 17–28 subsets) with 4.5 mm Gaussian smoothing. ADNI PET images were pre-processed and normalized to a uniform 6 mm full width at half maximum (FWHM). For CERMEP_IDB subjects, static FDG-PET scans were acquired on a Siemens Biograph mCT64. Images were reconstructed using 3D OP-OSEM with PSF and TOF modelling (12 iterations, 21 subsets), followed by Gaussian post-filtering (4 mm FWHM). Corrections for attenuation (CT-based), scatter, and normalization were applied during reconstruction. For Marseille subjects, images were acquired on a GE Discovery PET/CT, with a 6.2 mm axial resolution. Images were CT-corrected for attenuation and reconstructed into 47 contiguous 3.27 mm thick transverse sections using OSEM (5 iterations, 32 subsets).
For each patient, we retrieved a clinical T1-weighted anatomical MRI and an interictal FDG-PET scan acquired as close to the surgical date as possible. Due to the long inclusion period, acquisition parameters varied. Most T1-weighted scans (n = 37, 92.5%) featured isotropic or near-isotropic voxels (e.g. 1.0 × 1.0 × 1.0 mm3 or 1.0 × 1.0 × 1.2 mm3). The remaining scans (n = 3, 7.5%) had higher degrees of anisotropy (0.7 × 1.0 × 0.7 mm3 or 0.9 × 0.9 × 1.2 mm3). Patients followed a standardised clinical protocol involving a confirmed fasting state (glycemia < 180 mg/dL) and the injection of approximately 150 MBq of FDG. PET images were acquired on four systems: Siemens ECAT EXACT HR+ , Biograph HiRez PET-CT, Biograph TruePoint PET-CT (Siemens Healthineers AG, Erlangen, Germany), and GE Discovery MI 4 PET-CT (GE HealthCare, Chicago, IL, USA). Images were reconstructed using iterative algorithms (n = 18) or filtered back projection (n = 22), correcting for scatter, randoms, dead-time, radioactive decay, and attenuation (68Ge-based for the ECAT HR + system, CT-based for all others).
Image processing and quality control
FDG-PET images were co-registered to pre-operative T1-weighted scans, and grey matter probability maps were segmented using CAT12 v12.9 using a standardised pipeline (see Supplementary Table 1). We visually inspected all co-registration and segmentation outputs. We applied a strict quality control protocol, flagging subjects where meningeal or venous structures were misclassified as grey matter, as more subtle grey matter misclassifications could cause large false-positive asymmetries. The surgically resected volume was segmented after rigid co-registration of the post-operative imaging (MRI or CT scans) to pre-operative T1-weighted images by E.C.
ACAI calculation
An asymmetry index calculates the percentage difference between the PET activity in a voxel 
and its contralateral homologue 
. However, to correct for anatomical asymmetries, ACAI first calculates an intermediate, anatomy-corrected volumetric image (
). 
represents a local average of PET activity, weighted for grey matter probability. 
is calculated as a ratio of two convolutions, using a 9 mm FWHM Gaussian kernel.
 |
is the resulting anatomy-corrected value at the central voxel i.
is the convolution kernel (e.g., a 9 mm FWHM Gaussian kernel) centred at voxel i.
A is the fuzzy segmentation map for the tissue of interest (i.e., grey matter).
is the weight of the kernel (9 mm FWHM) at voxel j, centred at voxel i.
is the grey matter probability at voxel j.
is the original FDG-PET intensity at voxel j.
These ACAI calculations were performed in Montreal Neurological Institute (MNI) space. The final ACAI maps were then masked to exclude voxels with a grey matter probability < 0.25 and warped back to each subject's native space.
For each subject (patients and healthy controls), we determined the lowest ACAI threshold at which a cortical hypometabolic cluster (0.5, 1.0, 2.0, 5.0, 10.0 mL) remained detectable in subject space, using a Neuromorphometrics atlas-based parcellation. These values were then compared across cohorts, hemispheres, and segmentation quality.
ACAI threshold stability in healthy controls
Mean ACAI thresholds and threshold variance were compared across the cohorts to evaluate inter-cohort consistency. The impact of segmentation quality was assessed by comparing results from the full cohort to a subset containing only subjects with high-quality segmentations. Mean hemispheric thresholds were also compared to assess inter-hemispheric consistency. Finally, we evaluated threshold stability across lifespan by correlating ACAI thresholds with age in healthy controls.
Discriminative performance of ACAI for patient identification
To establish optimal thresholds for clinical validation, we defined hypometabolic volumes using an ACAI and cluster threshold that yielded the highest discriminative accuracy between patients and healthy controls. We selected the cluster size that yielded the highest discriminative performance (area under the curve [AUC]) in a receiver operating characteristic (ROC) analysis discriminating patients from controls. We based the corresponding ACAI threshold on the Youden's index. In addition, we also reported discriminative performance and corresponding thresholds across cohorts and hemispheres, using a cluster volume of 2.0 mL, at Youden’s index and at 95% specificity.
Clinical evaluation of ACAI in patients with FCD
For all subsequent analyses, we defined hypometabolic volume using the optimal threshold of -19 and cluster size > 2.0 mL based on the pooled healthy control cohort. Hypometabolic volume represents the total volume of cortical voxels within each cluster meeting these criteria, calculated in native subject space. We evaluated its ability to (i) lateralise the epileptogenic hemisphere, (ii) localise the surgical target, and (iii) predict seizure freedom at 24 months postoperatively.
To assess lateralisation, we compared the total ipsilateral versus contralateral hypometabolic volumes. To assess localisation, we quantified the number and distribution of hypometabolic clusters, determining how many were ipsilateral and how many were "localising", defined as having > 10% of their volume overlapping with the resection. The functional deficit zone often extends beyond the epileptogenic zone, and complete resection of the hypometabolic volume is not required for seizure freedom [20]. We applied this minimum threshold to reduce the inclusion of marginal overlaps that might result from a spatially extensive hypometabolic volume rather than true localisation. We further quantified spatial overlap by calculating the percentage of the total hypometabolic volume resected and the Dice coefficient, calculated as twice the volume of spatial overlap between the hypometabolic and resected regions divided by the sum of their individual volumes. This metric yields a value from 0 to 1, where 0 indicates an absence of spatial overlap and 1 indicates complete spatial congruence between the two volumes.
For outcome prediction, we first performed a cluster-based analysis to examine whether specific patterns, such as the presence of a localising cluster, no clusters, or only non-localising clusters, were associated with favourable outcomes. We then evaluated the percentage of resected hypometabolic volume as a predictor, treating it as both a continuous and a dichotomous (> 10% vs < 10%) variable. Patients without hypometabolic clusters were considered to have 0% resected hypometabolic volume. Finally, to test the robustness of our findings, we conducted sensitivity analyses using more permissive and stringent ACAI thresholds (- 4 to + 4 of ideal threshold), more permissive and stringent resected hypometabolic volumes (5–20%), and performed an analysis including all patients regardless of segmentation quality.
Statistical analysis
Statistical analyses were performed in Python, using scikit-learn (v1.5.0), SciPy (v1.13.1) and statsmodels (v0.14.2). We assessed threshold variability across control datasets using Levene’s test and then performed an ANOVA for assessing differences in means, followed by post-hoc Tukey's tests (family-wise error rate 0.05). Inter-hemispheric asymmetries were examined using paired t-tests. We calculated the Spearman correlation coefficient between subject age and threshold in the pooled healthy control cohort and within each individual cohort. To correct for multiple comparisons across cluster thresholds, we applied the Benjamini–Hochberg procedure to control the false discovery rate (FDR). Clusters were considered statistically significant if they met the FDR threshold (q < 0.05). For these significant clusters, the original, uncorrected p-values are reported.
We quantified discriminative performance via ROC analysis and AUC metrics. We reported sensitivities, specificities and thresholds for Youden’s index and > 95% specificity. AUCs were compared using bootstrapping (2000 samples). We compared percentage resected hypometabolic volumes in patients (Engel I vs Engel > I) using a Mann–Whitney U test. We assessed the association between dichotomized resected hypometabolic volume and seizure outcome using Fisher’s exact test. Comparably, we also assessed the association between cluster status (localising cluster vs. no clusters and vs. no localising clusters) and seizure freedom using Fisher’s exact test. To quantify the predictive value of resected hypometabolic volume, we conducted both univariable and exploratory multivariable logistic regression analyses, adjusting for MRI lesion status (one congruent lesion, normal MRI or multiple/non-congruent lesions) and FCD pathology, and corrected separately for PET visual assessment, acknowledging the limitations of sample size.
Results
Cohort characteristics
A total of 179 healthy control participants (76 males, 42.5%) were included from five distinct cohorts. These comprised two in-house cohorts, a younger in-house control group (Y_IHC; n = 26, median age 47.0 years, IQR: 24.3–52.8; 42.3% male), an older in-house control group (O_IHC; n = 20, median age 65.0 years, IQR: 54.8–69.5; 35.0% male), and three external cohorts from the ADNI (n = 44, median age 70.0 years, IQR: 64.8–73.0; 43.2% male), CERMEP_IDB (n = 37, median age 35.0 years, IQR: 27.0–46.0; 45.9% male) and Marseille databases (n = 52, median age 49.0 years, IQR: 35.0–62.8; 42.3% male).
The patient cohort comprised 40 patients (median age 29.5 years, IQR: 23.8–37.3; 55% male). The median epilepsy duration prior to resection was 22 years. Surgical resections were primarily performed in the temporal (n = 16, 40.0%) and frontal (n = 14, 35.0%) lobes. MRI findings were congruent in 22 cases (55%). At 24-month follow-up, 25 patients (62.5%) achieved seizure freedom (Engel Class I). Age did not significantly differ between patients with Engel Class I and Engel Class > I (median age 33 years, IQR: 23.2–42.5 vs median age 27.5 years, IQR: 24.5–32.0). Detailed clinical characteristics are summarized in Table 1.
Table 1.
Clinical and radiographic characteristics
| All patients (n = 40) | Patients with high-quality segmentations (n = 29) | |
|---|---|---|
| Sex | N (%) | N (%) |
| Male | 22 (55.0%) | 18 (62.1%) |
| Female | 18 (45.0%) | 11 (37.9%) |
| Age at seizure onset and resection | Median (IQR) | Median (IQR) |
| Age at Seizure Onset (years) | 8.0 [4.0—15.0] | 12.0 [4.0—17.0] |
| Age at Surgical Resection (years) | 29.5 [23.8—37.3] | 29.0 [24.0—37.0] |
| ASM Count | Median (IQR) | Median (IQR) |
| Pre-Surgery | 3.0 [2.0—3.0] | 3.0 [2.0—3.0] |
| 24-Months Post-Op | 2.0 [1.0—3.0] | 2.0 [1.0—3.0] |
| Surgery Location | N (%) | N (%) |
| Temporal lobe | 16 (40.0%) | 12 (41.4%) |
| Frontal lobe | 14 (35.0%) | 7 (24.1%) |
| Occipital lobe | 5 (12.5%) | 5 (17.2%) |
| Parietal lobe | 4 (10.0%) | 4 (13.8%) |
| Insular lobe | 1 (2.5%) | 1 (3.5%) |
| MRI Findings | N (%) | N (%) |
| Congruent MRI lesion | 22 (55.0%) | 16 (55.2%) |
| Normal MRI | 15 (37.5%) | 11 (37.9%) |
| Non-congruent or multiple lesions | 3 (7.5%) | 2 (6.9%) |
| PET lateralisation* | N (%) | N (%) |
| Ipsilateral to resected side | 27 (67.5%) | 19 (65.5%) |
| Contralateral to resected side | 1 (2.5%) | 1 (3.5%) |
| Bilateral | 7 (17.5%) | 6 (20.7%) |
| Normal | 5 (12.5%) | 3 (10.3%) |
| Engel Outcome at 24 Months | N (%) | N (%) |
| Class I | 25 (62.5%) | 16 (55.2%) |
| Class II | 9 (22.5%) | 9 (31.0%) |
| Class III | 4 (10.0%) | 3 (10.3%) |
| Class IV | 2 (5.0%) | 1 (3.4%) |
| FCD type | N (%) | N (%) |
| I | 11 (27.5%) | 9 (31.0%) |
| IIa | 16 (40.0%) | 14 (48.3%) |
| IIb | 13 (32.5%) | 6 (20.7%) |
ASM = anti-seizure medication; FCD = focal cortical dysplasia; IQR = interquartile range; MRI = magnetic resonance imaging; PET = positron emission tomography
* Qualitative PET findings in relation to the surgical side
ACAI threshold stability across healthy control cohorts
We first assessed ACAI threshold variability across the full control cohort (n = 179). Significant inter-dataset variance was observed for smaller cluster sizes, with Levene’s test indicating heterogeneity at 0.5 mL (P = 0.004) (Fig. 1A). Visual inspection revealed that segmentation errors, particularly misclassified meninges and venous structures, contributed substantially to this variability (Fig. 2B).
Fig. 1.
ACAI threshold distribution in healthy controls and diagnostic performance. (A) Boxplots illustrating the distribution of the least negative ACAI threshold required to detect at least one hypometabolic cluster in healthy controls. This is shown for three different minimum cluster sizes (0.5, 2.0, and 10.0 mL), for both all controls (‘All’, n = 179) and only controls with good segmentations (‘Good’, n = 131). The letters A, B, C, D, and E correspond respectively to the ADNI cohort, in-house younger cohort, in-house older cohort, CERMEP_IDB cohort and Marseille cohort. Each point represents an individual subject. Significant Post-hoc Tukey tests are shown with *. (B) ROC curves demonstrating the performance of the ACAI threshold in distinguishing patients with FCD (n = 29) from healthy controls (n = 131) with good segmentations for three different minimum cluster sizes (0.5, 2.0, and 10.0 mL). AUCs are provided for each of the five control cohorts (A, B, C, D, E) indicating optimal discriminative ability at a threshold of 2.0 mL. ACAI = anatomy-corrected asymmetry index; ADNI = Alzheimer’s Disease Neuroimaging; All = all healthy control subjects (n = 179); AUC = area under the curve; FCD = focal cortical dysplasia; Good = good segmentations; Segm. = segmentations; ROC = receiver operating characteristic
Fig. 2.
Segmentation artefacts and inter-hemispheric differences. This figure illustrates (A) an example of a segmentation error (red arrow) causing an abnormally low ACAI cluster (< -30) in a healthy control (red volume), (B) a heat map (percentage of voxels with ACAI values < -30) showing typical regions of lower clusters due to segmentation errors, including the mis‑segmentation of the transverse sinus, and (C) mean ACAI values (controls with good segmentations n = 131) showing consistent higher values in the right temporal and frontal lobe. The color scales indicate percentages (B) or ACAI values (C). ACAI = anatomy-corrected asymmetry index; L = left; R = right
After excluding healthy controls with segmentation artefacts (n = 48), variance across datasets became homogeneous in the remaining 131 healthy controls (Levene’s test, p > 0.14) (Fig. 1A).
We excluded 11 patients specifically due to meningeal or venous structures misclassified as grey matter. We restricted subsequent analyses to 131 healthy controls and 29 patients. For small-to-medium clusters (0.5–2.0 mL), mean ACAI thresholds were consistent across datasets (ANOVA, p > 0.19). Larger clusters (5.0–10.0 mL) still showed small (< 2.6) but significant differences in mean ACAI thresholds across datasets (5.0 mL: p = 0.0015; 10.0 mL: p = 0.0075), with post hoc Tukey's tests revealing significantly lower ACAI thresholds in the older ADNI cohort compared to in-house controls. In the pooled healthy control cohort, we observed a weak but significant negative correlation between age and the ACAI threshold (ρ = -0.17, P = 0.049). This remained significant within two individual cohorts (ADNI: ρ = -0.59, P = 0.002; Marseille: ρ = -0.30, P = 0.048).
Across all cluster volumes, ACAI thresholds were consistently lower in the left hemisphere (P values ranged from < 0.001 to < 0.05), with mean absolute differences ranging from 2.3 (0.5 mL) to 3.6 (10.0 mL) (Supplementary Fig. 1). Hemispheric asymmetries are visualized in Fig. 2A.
Discriminative performance of ACAI for patient identification
ACAI thresholds in patients were lower than in controls across all cluster volume thresholds. Highest accuracy occurred at a cluster threshold of 2.0 mL, with AUCs of 0.87 (ADNI), 0.85 (Y_IHC), 0.79 (O_IHC), 0.86 (CERMEP_IDB) and 0.87 (Marseille) (Fig. 1B). At this cluster size, Youden’s index identified optimal thresholds of -19 (ADNI, O_IHC, CERMEP_IDB) and -17 (Y_IHC, Marseille), yielding sensitivities of 69–72% and specificities of 93–100%. When all healthy controls were pooled, highest AUC (0.85) was obtained at 2.0 mL and optimal threshold (Youden’s index) was -19. This threshold achieved a sensitivity of 69% (20 out of 29 FCD patients had at least one cluster) and a specificity of 97% (3% of healthy controls still had at least one cluster). Accuracy (AUC) was not significantly higher compared to other tested cluster thresholds (0.5, 1.0, 5.0, 10.0 mL).
When comparing patients by operated hemisphere (left n = 14, right n = 15) to ipsilateral hemispheres in controls, Youden's index identified optimal thresholds of -17 for the left hemisphere (64.3% sensitivity, 96.9% specificity) and -11 for the right hemisphere (86.7% sensitivity, 76.3% specificity). Specificity exceeded 95% at thresholds of -17 (64.3% sensitivity, 96.9% specificity) for the left hemisphere and -19 (60.0% sensitivity, 98.5% specificity) for the right.
For all subsequent analyses, we defined hypometabolic volume using the optimal threshold of -19 and cluster size > 2.0 mL based on the pooled healthy control cohort. We performed sensitivity analyses at thresholds of -15, -17, -21, -23 (Supplementary Table 2). Figure 1B shows ROC curves for different cluster sizes (0.5, 2.0 and 10.0 mL).
Localisation and lateralisation of ACAI clusters in patients with FCD
Using the predefined ACAI thresholds (< -19, > 2.0 mL), 20 of the 29 patients with high-quality segmentations had detectable hypometabolic clusters. In these 20 patients, a total of 32 clusters were detected (median 1 cluster per patient, IQR: 1–2, Range: 1–5). Nine patients had one single localising cluster. Twenty-four clusters were ipsilateral to the resection zone, and eight clusters were contralateral. Overall, 17 of 32 clusters (53.1%) were localising (> 10% resected). Localising clusters showed substantial overlap, with a median resection volume of 47.2% (IQR: 35.9%–75.8%; Range: 25.3%–88.9%). All non-localising clusters (n = 15) showed 0% overlap with the resection volume.
At the per-patient level, the median ipsilateral hypometabolic volume was 6.4 mL (IQR: 3.3–9.2) compared to a median contralateral volume of 0.0 mL (IQR: 0.0–0.77). Across the entire cohort (n = 29), the median resected hypometabolic volume was 29.9% (IQR: 0.0–48.3), with a median Dice coefficient of 0.25 (IQR: 0.0–0.41).
Localising ACAI clusters predict seizure freedom
Patients were stratified into three groups: nine (31.0%) had no clusters, 17 (58.6%) had at least one localising cluster (overlapping > 10% with resection), and three (10.3%) had only non-localising clusters. Patients with at least one localising cluster had significantly better outcomes than those with no clusters (76.5% [13/17] vs. 22.2% [2/9] Engel I, Fisher's exact test P = 0.014). Seizure freedom rates were 76.5% (13/17) in the localising group compared to 33.3% (1/3) in the non-localising group, a difference that did not reach statistical significance (P = 0.20). Notably, the most favourable prognosis was observed in patients with a single localising cluster, of whom eight out of nine (88.9%) achieved seizure freedom.
We then assessed the resected hypometabolic volume as a predictor of outcome for all 29 patients. Patients with a good outcome (Engel I, n = 16) had a significantly larger resected hypometabolic volume percentage (median 44.9%, IQR: 27.1–71.3) compared to those with a poor outcome (Engel > I, n = 13) (median 0.0%, IQR: 0.0–35.5) (Mann–Whitney U test P = 0.011) (Fig. 3). Percentage resected hypometabolic volume remained an independent predictor of seizure freedom after correcting for MRI status and FCD pathology (OR per 10% increase: 1.40, 95% CI: 1.01–1.92, P = 0.042). In a separate model adjusting for qualitative PET assessment, the odds ratio per 10% increase in resected volume was 1.36 (95% CI: 0.99-1.88, P = 0.057). When dichotomized, resecting at least 10% of the hypometabolic volume was associated with significantly better outcomes (76.5% [13/17] vs. 25.0% [3/12], Fisher's exact test P = 0.01).
Fig. 3.
Resected hypometabolic volume and outcome. Patients who achieved a good outcome had a significantly higher percentage of their hypometabolic volume resected compared to those with a poor outcome (Mann–Whitney U test, P = 0.011)
Among the nine patients without ACAI clusters in the high-quality segmentation cohort, all had abnormal qualitative PET scans (four with bilateral hypometabolism, four ipsilateral), yet only 22.2% (2/9) achieved seizure freedom. These two patients with good outcomes only displayed a cluster within the resection zone at a threshold of < -16. Conversely, three patients had normal PET scans but positive ACAI clusters, and two of these (66.7%) achieved good outcomes. One patient displayed hypometabolism contralateral to the surgical resection zone on visual assessment. ACAI identified one right-sided, contralateral cluster. This patient did not achieve seizure freedom (Engel Class III). Four examples of patients with ACAI clusters are shown in Fig. 4.
Fig. 4.
Clinical examples of ACAI-defined hypometabolism. The resection zone is depicted in green, the ACAI cluster in red. Patient A illustrates an MRI-positive FCD case (blue arrow) where qualitative PET was initially interpreted as normal. ACAI detected a single cluster concordant with the lesion, and the patient achieved seizure freedom. Patient B had a lesion suggestive for FCD in the anterior temporal lobe on MRI (blue arrow), concordant EEG and ictal SPECT findings, and a marked hypometabolic pattern on visual PET analysis. The patient achieved seizure freedom after resection. However, the patient was a false negative where ACAI < -19 failed to detect any clusters at the standard threshold and only showed a localising cluster at < -16 (visualized here). This case demonstrates that stringent ACAI thresholds can exclude clinically relevant regions. Patient C, an MRI-negative, PET-positive TLE case, showed one cluster with clear overlap. The patient achieved seizure freedom. Patient D had only subtle MRI changes, but a clear zone of hypometabolism in the mesial occipital cortex. A large proportion of the ACAI-defined hypometabolic volume was resected, yet seizure freedom was not achieved at 24 months. MRI revealed only subtle FCD. The patient was readmitted four years later for a new work-up. Intracranial EEG later identified a seizure onset zone adjacent to the unresected posterior ACAI cluster (yellow zone and yellow arrow), and reoperation targeting this region ultimately resulted in seizure freedom. This case underscores the fact that unresected ACAI clusters may correspond to residual epileptogenic tissue. All shown images are pre-operative scans. ACAI = Anatomy-Corrected Asymmetry Index; EEG = electroencephalography; FCD = focal cortical dysplasia; FDG = fluorodeoxyglucose; FLAIR = fluid-attenuated inversion recovery; MRI = magnetic resonance imaging; PET = positron emission tomography; SPECT = single-photon emission computed tomography; TLE = temporal lobe epilepsy
To evaluate the robustness of our findings, we repeated the outcome prediction analysis on the full cohort (N = 40). This analysis included 11 patients previously excluded due to misclassified meninges or venous structures. The percentage of resected hypometabolic volume remained a significant predictor of seizure freedom (Mann–Whitney U test P = 0.02) (Supplementary Fig. 2). Results remained comparable across ACAI thresholds from -15 to -23 (Supplementary Table 2) and volume overlap thresholds from 5 to 20% (Supplementary Table 3). The highest spatial overlap (Dice coefficient) was achieved at a threshold of -15, while the maximum resected hypometabolic volume was observed at -20 (Supplementary Fig. 3). To evaluate the trade-off between diagnostic specificity and surgical sensitivity, we assessed the percentage of patients with any resected overlap across a continuous range of thresholds (-5 to -30) (Supplementary Fig. 4). All seizure-free (Engel I) patients with good segmentations (n = 16) exhibited resected hypometabolic volume at a threshold of -16.
Discussion
In this study, we evaluated FDG-PET ACAI across multiple, multicentric healthy control cohorts. Other asymmetry approaches evaluated smaller cohorts and focused primarily on lateralisation or exclusively on patients with favourable outcomes [12, 21, 22]. This multicentric evaluation enabled us to establish normative thresholds and assess inherent technique limitations. Furthermore, ACAI improves upon standard indices by integrating prior structural information from MRI segmentation to minimize the influence of anatomical variations [11]. This clinical validation allowed us to assess ACAI’s generalisability and establish a standardised framework.
The reliability of ACAI is directly dependent on the quality of the underlying anatomical segmentation. Misclassification of non-brain tissue such as meninges or vessels led to false-positive clusters, especially at smaller volumes (0.5–1 mL). We conservatively excluded all subjects with visible grey matter misclassifications to ensure robust threshold definition. This strict criterion led to high exclusion rates (27% of controls, 28% of patients). While gross segmentation errors render ACAI unusable, a reliable implementation requires a high-quality 3D T1-weighted MRI alongside visual quality control of PET-MRI co-registrations, segmentation maps, and ACAI clusters to exclude artefacts and retain valid findings.
After excluding patients with poor segmentations, inter-cohort differences were small, despite differences in age, scanner acquisition protocols, and attenuation correction methods (ZTE versus CT-based). Inter-hemispheric differences were more pronounced, likely attributable to a combination of true metabolic asymmetries and technical factors. Inherent metabolic asymmetries are well documented in language and visuospatial regions [23]. We observed a weak negative correlation between age and ACAI thresholds, aligning with observed age-related progression of these metabolic asymmetries [23]. This might be clinically relevant in rare cases of older patients undergoing epilepsy surgery, where age-related structural and metabolic asymmetries might require stricter thresholds. Structurally, ACAI does not fully account for neuroanatomical shifts like the Yakovlevian torque, which involves anterior right frontal and posterior left occipital protrusions [24]. Asymmetrical regional cortical thickness and surface area also introduce partial volume effects [25]. While ACAI reduces these effects, future frameworks could integrate explicit partial volume correction algorithms. Together, these factors likely drive the consistent rightward elevation of mean ACAI values observed in temporal and frontal regions (Fig. 2C).
However, these variations did not meaningfully degrade the method's diagnostic performance. Across all control datasets, values of -19 or -17 consistently yielded specificities of 93–100% (AUC 0.79–0.87). While inter-hemispheric Youden's index thresholds differed, the thresholds yielding > 95% specificity were highly comparable (left: -17, right: -19). Given this high concordance, and as -19 was the optimal threshold for the pooled cohorts, we defined hypometabolic volume using this unified threshold of -19 and a cluster size > 2.0 mL for all subsequent analyses. Nevertheless, future studies utilizing larger validation cohorts could explore whether adopting hemisphere-specific thresholds might further increase diagnostic accuracy.
After establishing a normative framework, we then tested its clinical utility in our FCD cohort. The extent of resected ACAI-defined hypometabolic volume was a significant predictor of surgical outcome, independent of FCD pathology or MRI lesion status. Patients with a resection of at least 10% of the hypometabolic volume had significantly higher rates of seizure freedom (76.5% vs. 25.0%, P = 0.01), and this predictive value was greatest for patients with a single, localising ACAI cluster (88.9% seizure freedom). These findings align with recent reports showing that localising FDG-PET hypometabolism predicts better surgical outcomes than diffuse or absent hypometabolism, independently of MRI [26–28].
Furthermore, our findings show a potential additional value of ACAI relative to traditional qualitative assessment. In our cohort, nine patients deemed abnormal on visual PET review had no detectable ACAI clusters. This group had a poor prognosis, with only 22.2% (2/9) achieving seizure freedom. This discrepancy could arise from bilaterally symmetric hypometabolism, which an asymmetry index cannot detect by design. Alternatively, the visual assessment may have identified subtle pathological or physiological metabolic asymmetries that did not meet the stringent cluster threshold. Conversely, ACAI identified positive clusters in three patients with initially normal visual PET scans; two of these (66.7%) achieved good outcomes. ACAI might thus provide objective data that can complement the standard clinical read.
When implementing ACAI, one might use a different threshold selection to balance sensitivity and specificity for surgical planning. While a threshold of -19 was optimized for discrimination between patients and controls, a more permissive threshold may be preferable in a multimodal clinical context to maximize sensitivity and ensure no potential zone of interest is missed during surgical planning [29]. We therefore performed sensitivity analyses demonstrating comparable results using slightly different thresholds (-15 to -23) and found that all patients with high-quality segmentations and good outcome exhibited a cluster overlapping with the resection zone at -16. Future studies in larger datasets should define optimal thresholds that maximize sensitivity for the resection zone in seizure-free patients.
Beyond threshold selection, ACAI-defined clusters should be integrated with other presurgical data to guide surgical strategy [29]. A single cluster concordant with electroclinical and structural findings provides strong evidence for a surgical hypothesis and may inform a more targeted resection approach.
However, the functional deficit zone frequently extends beyond the epileptogenic zone resulting in multiple clusters with only partial resection (median 29.9% in our cohort) [20]. In such cases, concordance of one cluster with independent localising data can help consolidate competing hypotheses [26]. Conversely, the discovery of spatially discordant clusters should prompt a thorough re-evaluation of all data. A third scenario is the absence of any detectable ACAI clusters. As noted above, this finding was associated with a poor prognosis even when visual PET interpretation suggested subtle abnormalities. This demonstrates how ACAI might complement visual interpretation and could directly inform surgical decision-making to improve outcomes.
The primary limitations of this study are its retrospective, single-center and modestly sized patient cohort, limiting generalisability of our findings in patients with FCD. This modest sample size also prevented a clear comparison between patients with no hypometabolic clusters and those with only non-localising clusters and impacted our ability to discern effects of covariates such as FCD type, MRI findings, qualitative PET findings, age and ASM. Second, the reliability of ACAI depends on the quality of the anatomical segmentation. In this study, we excluded 11 patients and 48 healthy controls due to artefacts associated with clinical-grade, sometimes non-isotropic, MRI scans. We utilized a standard CAT12 pipeline which, while widely validated, can be prone to misclassifying meningeal or vascular structures as grey matter. We deliberately avoided optimizing segmentation parameters for individual subjects to avoid introduction of systematic bias. Although a sensitivity analysis incorporating these excluded patients confirmed the predictive value of the method, the precision of the resection overlap decreased in this unfiltered cohort. Clinical implementation requires a rigorous visual quality control step, where segmentation maps are visually assessed alongside ACAI maps. Our study could not definitively disentangle the effects of age from scanner/protocol differences, as seen in the older ADNI cohort which showed lower mean thresholds. While this could partly reflect scanner differences, our exploratory analysis revealed a significant negative correlation between age and the ACAI threshold within the ADNI cohort. This suggests that physiological brain ageing introduces metabolic or structural asymmetries that necessitate stricter thresholds. Future studies could investigate whether an age-adjusted threshold could improve diagnostic accuracy in elderly individuals.
Future research should focus on prospective validation of ACAI in larger, multicenter and multimodal cohorts to establish its generalisability across different clinical settings and epileptic lesions other than FCD, and its impact on surgical decision-making and planning.
Conclusion
FDG-PET ACAI provides a standardised method for identifying hypometabolic regions in patients with FCD and is associated with epilepsy surgery outcome. Clinical implementation requires high-quality anatomical segmentation or careful exclusion of artefactual clusters.
Supplementary Information
Below is the link to the electronic supplementary material.
Acknowledgements
We gratefully acknowledge the CERMEP-Imagerie du Vivant (Lyon, France) for providing the CERMEP-IDB-MRXFDG dataset. We also thank the Department of Nuclear Medicine and the Department of Neuroradiology at Timone Hospital (AP-HM) and Aix-Marseille University (Marseille, France) for sharing the Marseille healthy control dataset. We thank the investigators involved in the creation and harmonization of these open-access databases. We are also grateful to the volunteers who participated in these studies.
Author contributions
All authors contributed to the study conception and design. Material preparation and analysis were performed by Jeroen Gijs, Evy Cleeren, and Jaiver Macea. The original Anatomy-Corrected Asymmetry Index was developed by Patrick Dupont and implemented in this study by Jeroen Gijs.
Funding
This research was supported by the Research Foundation Flanders (FWO T000423N). Data collection and sharing for ADNI is funded by the National Institute on Aging (National Institutes of Health Grant U19AG024904). Tom Theys is Senior Clinical Investigator of FWO Flanders (FWO 1830722N). Greet Vanderlinden is funded by a FWO junior postdoctoral fellowship (FWO 1292326N).
Data availability
The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request. Code for calculating ACAI images and thresholding these images can be found at https://github.com/jeroengijs/ACAI.
Declarations
Ethical approval
This study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the Ethics Committee UZ/KU Leuven.
Consent to participate
All participants in the in-house healthy control cohorts provided written informed consent. For the retrospective patient cohort, the Ethics Committee approved the use of data based on the institutional opt-out consent procedure.
Consent to publish
The authors affirm that human research participants provided informed consent for publication of the data.
Competing interests
The authors report no competing interests.
Footnotes
*Data used in preparation of this article were obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this report. A complete listing of ADNI investigators can be found at: http://adni.loni.usc.edu/wp-content/uploads/how_to_apply/ADNI_Acknowledgement_List.pdf.
Wim Van Paesschen and Karolien Goffin contributed equally to this work.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
- 1.Tassi L, Garbelli R, Colombo N, et al. Type I focal cortical dysplasia: surgical outcome is related to histopathology. Epileptic Disord. 2010;12(3):181–91. 10.1684/epd.2010.0327. [DOI] [PubMed] [Google Scholar]
- 2.Kim DW, Lee SK, Chu K, et al. Predictors of surgical outcome and pathologic considerations in focal cortical dysplasia. Neurology. 2009;72(3):211–6. 10.1212/01.WNL.0000327825.48731.C3. [DOI] [PubMed] [Google Scholar]
- 3.Tassi L, Colombo N, Garbelli R. Focal cortical dysplasia: neuropathological subtypes, EEG, neuroimaging and surgical outcome. Brain. 2002;125(8):1719–32. 10.1093/brain/awf175. [DOI] [PubMed] [Google Scholar]
- 4.Najm IM, Sarnat HB, Blümcke I. Review: the international consensus classification of focal cortical dysplasia – a critical update 2018. Neuropathol Appl Neurobiol. 2018;44(1):18–31. 10.1111/nan.12462. [DOI] [PubMed] [Google Scholar]
- 5.Spitzer H, Ripart M, Whitaker K, et al. Interpretable surface-based detection of focal cortical dysplasias: a multi-centre Epilepsy Lesion Detection study. Brain. 2022;145(11):3859–71. 10.1093/brain/awac224. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Wang ZI, Jones SE, Jaisani Z, et al. Voxel-based morphometric magnetic resonance imaging (MRI) postprocessing in MRI-negative epilepsies. Ann Neurol. 2015;77(6):1060–75. 10.1002/ana.24407. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.David B, Kröll-Seger J, Schuch F, et al. External validation of automated focal cortical dysplasia detection using morphometric analysis. Epilepsia. 2021;62(4):1005–21. 10.1111/epi.16853. [DOI] [PubMed] [Google Scholar]
- 8.Jin SO, Mérida I, Stavropoulos I, et al. Characterisation of a novel [18F]FDG brain PET database and combination with a second database for optimising detection of focal abnormalities, using focal cortical dysplasia as an example. EJNMMI Res. 2023;13(1):98. 10.1186/s13550-023-01023-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Didelot A, Mauguière F, Redouté J, et al. Voxel-based analysis of asymmetry index maps increases the specificity of 18F-MPPF PET abnormalities for localizing the epileptogenic zone in temporal lobe epilepsies. J Nucl Med. 2010;51(11):1732–9. 10.2967/jnumed.109.070938. [DOI] [PubMed] [Google Scholar]
- 10.Ohta Y, Nariai T, Ishii K, et al. Voxel- and ROI-based statistical analyses of PET parameters for guidance in the surgical treatment of intractable mesial temporal lobe epilepsy. Ann Nucl Med. 2008;22(6):495–503. 10.1007/s12149-008-0140-5. [DOI] [PubMed] [Google Scholar]
- 11.Zhou L, Dupont P, Baete K, Van Paesschen W, Van Laere K, Nuyts J. Detection of inter-hemispheric metabolic asymmetries in FDG-PET images using prior anatomical information. Neuroimage. 2009;44(1):35–42. 10.1016/j.neuroimage.2008.07.047. [DOI] [PubMed] [Google Scholar]
- 12.Aslam S, Damodaran N, Rajeshkannan R, Sarma M, Gopinath S, Pillai A. Asymmetry index in anatomically symmetrized FDG-PET for improved epileptogenic focus detection in pharmacoresistant epilepsy. J Neurosurg. 2023;138(3):828–36. 10.3171/2022.6.JNS22717. [DOI] [PubMed] [Google Scholar]
- 13.Jack CR Jr, Bernstein MA, Fox NC, et al. The Alzheimer’s Disease Neuroimaging Initiative (ADNI): MRI methods. J Magn Reson Imaging. 2008;27(4):685–91. 10.1002/jmri.21049. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.van Aalst J, Jennen L, Demyttenaere K, et al. Twelve-week yoga vs. aerobic cycling initiation in sedentary healthy subjects: a behavioral and multiparametric interventional PET/MR study. Front Psychiatry. 2021;12:739356. 10.3389/fpsyt.2021.739356. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Delva A, Michiels L, Koole M, Van Laere K, Vandenberghe W. Synaptic damage and its clinical correlates in people with early Huntington disease: a PET study. Neurology. 2022;98(1):e83-94. 10.1212/WNL.0000000000012969. [DOI] [PubMed] [Google Scholar]
- 16.Mérida I, Jung J, Bouvard S, et al. CERMEP-IDB-MRXFDG: a database of 37 normal adult human brain [18F] FDG PET, T1 and FLAIR MRI, and CT images available for research. EJNMMI Res. 2021;11(1):91. 10.1186/s13550-021-00830-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Engel J Jr, Van Ness PC, Rasmussen TB, Ojemann LM. Outcome with respect to epileptic seizures. In: Engel J, editor. Surgical Treatment of the Epilepsies. 2nd ed. New York: Raven Press; 1993. p. 609–21. [Google Scholar]
- 18.Palmini A, Najm I, Avanzini G, et al. Terminology and classification of the cortical dysplasias. Neurology. 2004;62(6 Suppl 3):S2-8. 10.1212/01.WNL.0000114507.30388.7E. [DOI] [PubMed] [Google Scholar]
- 19.Blümcke I, Thom M, Aronica E, et al. The clinicopathologic spectrum of focal cortical dysplasias: a consensus classification proposed by an ad hoc Task Force of the ILAE Diagnostic Methods Commission. Epilepsia. 2011;52(1):158–74. 10.1111/j.1528-1167.2010.02777.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Lüders HO, Najm I, Nair D, Widdess-Walsh P, Bingman W. The epileptogenic zone: general principles. Epileptic Disord. 2006;8(Suppl 2):S1-9. 10.1684/j.1950-6945.2006.tb00204.x. [PubMed] [Google Scholar]
- 21.Kini LG, Thaker AA, Hadar PN, et al. Quantitative [18]FDG PET asymmetry features predict long-term seizure recurrence in refractory epilepsy. Epilepsy Behav. 2021;116:107714. 10.1016/j.yebeh.2020.107714. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Aslam S, Rajeshkannan R, Sandya CJ, Sarma M, Gopinath S, Pillai A. Statistical asymmetry analysis of volumetric MRI and FDG PET in temporal lobe epilepsy. Epilepsy Behav. 2022;134:108810. 10.1016/j.yebeh.2022.108810. [DOI] [PubMed] [Google Scholar]
- 23.Willis MW, Ketter TA, Kimbrell TA, et al. Age, sex and laterality effects on cerebral glucose metabolism in healthy adults. Psychiatry Res. 2002;114(1):23–37. 10.1016/S0925-4927(01)00126-3. [DOI] [PubMed] [Google Scholar]
- 24.Kuo F, Massoud TF. Structural asymmetries in normal brain anatomy: a brief overview. Ann Anat. 2022;241:151894. 10.1016/j.aanat.2022.151894. [DOI] [PubMed] [Google Scholar]
- 25.Luders E, Narr KL, Thompson PM, Rex DE, Jancke L, Toga AW. Hemispheric asymmetries in cortical thickness. Cereb Cortex. 2006;16(8):1232–8. 10.1093/cercor/bhj064. [DOI] [PubMed] [Google Scholar]
- 26.Courtney MR, Antonic-Baker A, Chen Z, et al. Association of localizing 18F-FDG-PET hypometabolism and outcome following epilepsy surgery: systematic review and meta-analysis. Neurology. 2024;102(9):e209304. 10.1212/WNL.0000000000209304. [DOI] [PubMed] [Google Scholar]
- 27.Guo K, Cui B, Shang K, et al. Assessment of localisation accuracy and postsurgical prediction of simultaneous 18F-FDG PET/MRI in refractory epilepsy patients. Eur Radiol. 2021;31(9):6974–82. 10.1007/s00330-021-07738-8. [DOI] [PubMed] [Google Scholar]
- 28.Sainburg LE, Hoang J, Doss DJ, et al. Surgical targeting of lateralised 18F-fluorodeoxyglucose positron emission tomography hypometabolism relates to long-term epilepsy surgery outcomes. Epilepsia. 2025;66(8):2816–29. 10.1111/epi.18402. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Lado FA, Ahrens SM, Riker E, et al. Guidelines for specialized epilepsy centers. Neurology. 2024;102(4):e208087. 10.1212/WNL.0000000000208087. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request. Code for calculating ACAI images and thresholding these images can be found at https://github.com/jeroengijs/ACAI.