Electroencephalography-functional magnetic resonance imaging using a familial temporal lobe epilepsy model in cats: an exploratory methodological pilot study Authors: Takayuki Miura, Rikako Asada, Daisuke Hasegawa Published: 2026 ABSTRACT AbstractBackgroundReliable methods for identifying the epileptogenic zone (EZ) in veterinary medicine remain limited. Although intracranial electroencephalography (EEG) is considered the most dependable approach for EZ localization, noninvasive simultaneous EEG and functional magnetic resonance imaging (EEG-fMRI) has been applied in human medicine as an alternative.Hypothesis/ObjectivesDevelop and evaluate EEG-fMRI methodology for clinical application in veterinary medicine and evaluate the methodological feasibility of EEG-fMRI using cats with familial spontaneous epilepsy (FSECs).AnimalsTen cats with clear interictal epileptiform discharges on scalp EEG were selected from an FSEC colony with previously suggested amygdalo-hippocampal involvement.MethodsEach cat underwent 3 scalp EEG and EEG-fMRI sessions. After preprocessing, blood oxygen level-dependent (BOLD) activation maps were generated. Primary analyses evaluated amygdalo-hippocampal activation using small-volume correction (SVC) and group-level analysis.ResultsThe BOLD activation was detected in all sessions at an exploratory threshold, but no activation survived correction for multiple comparisons. The SVC analysis identified amygdalo-hippocampal activation in 16/30 sessions. Group-level analysis identified the largest cluster extending from the left parahippocampal gyrus to the hippocampal body.Conclusions and clinical importanceThis exploratory pilot study showed that EEG-fMRI could be technically implemented in FSECs, but individual-level EZ localization remains insufficient for clinical use. Further optimization, data from surgical candidates, and multimodal comparisons are needed. Introduction Epilepsy surgery requires accurate identification of the epileptogenic zone (EZ).1 In human medicine, multimodal noninvasive evaluations, including seizure semiology, long-term video electroencephalography (VEEG), neuropsychological testing, structural magnetic resonance imaging (MRI), interictal fluorodeoxyglucose positron emission tomography (FDG-PET), magnetoencephalography, and ictal single-photon emission computed tomography are used. When these findings are inconclusive or deep foci are suspected, invasive electrocorticography (ECoG) or stereotactic electroencephalography (SEEG) is required.2,3 In veterinary medicine, epilepsy surgery has recently emerged.4–9 However, reliable EZ localization remains challenging despite advances in surgical techniques. Presurgical evaluations for dogs and cats include neurologic examination, scalp EEG, ictal video-based semiology, long-term VEEG, structural MRI, interictal FDG-PET, and intracranial EEG, but evidence for these modalities remains limited and diagnostic accuracy requires improvement.10 Intracranial VEEG monitoring is considered the most reliable method for EZ identification in human medicine, with electrode placement often using navigation systems or robotic guidance.11,12 However, comparable techniques are rarely available in veterinary medicine,3 and stereotactic planning is prone to placement error. Although intracranial EEG has been reported in freely moving dogs and cats, its feasibility is constrained by motion artifacts, device-related complications, costs, ethical considerations, and owner acceptance. Moreover, invasive recordings such as ECoG and SEEG sample only restricted brain regions, making accurate electrode placement dependent on precise noninvasive presurgical localization. Electroencephalography–functional magnetic resonance imaging (EEG-fMRI), introduced in human medicine in the 1990s,13,14 combines the temporal resolution of EEG with the spatial resolution of blood oxygenation level-dependent (BOLD) fMRI. Electroencephalography–functional magnetic resonance imaging can visualize epileptogenic activity from superficial to deep structures. Unlike conventional task-based fMRI, EEG-fMRI treats interictal epileptiform discharges (IEDs) as event-related stimuli and identifies associated BOLD responses. This technique has demonstrated concordance with scalp EEG and intracranial EEG and has shown value as a presurgical tool in humans with epilepsy.15–18 To our knowledge, EEG-fMRI has not been reported in dogs or cats. Establishing this modality may improve noninvasive localization, guide invasive electrode placement, and decrease reliance on invasive testing. Our pilot study aimed to evaluate the methodological feasibility of EEG-fMRI in veterinary medicine. Using cats with familial spontaneous epilepsy (FSECs) with previously suggested amygdalo-hippocampal involvement,12,19–21 we evaluated whether simultaneous EEG-fMRI could be technically implemented using MRI-compatible EEG devices and a clinically applicable immobilization protocol, whether interpretable in-scanner EEG recordings could be obtained after artifact correction, and whether IED-related BOLD activation could be detected within this predefined region. Materials and methods Animals Ten FSECs were included based on previous reports indicating that the amygdalo-hippocampal region represents the presumed EZ.12,19–21 Details of seizure history and EEG findings are provided in Table 1 and the Appendix. Table 1Subject information.Cat #Age (Y)SexBody weight (kg)Seizure frequencySession #IED count (out-of-scanner EEG)Top 3 regions (out-of-scanner EEG)IED count (in-scanner EEG)Top 3 regions (in-scanner EEG) Cat 1 4F2.73/year145 (4.5/min)LT RC LC114 (9.5/min)LO RO LF275 (7.5/min)LF LC LT64 (5.3/min)LF RF LO341 (4.1/min)LF RF LT86 (7.2/min)LO LF RO Cat 2 1M3.71-2/year132 (3.2/min)LF LT RT105 (8.75/min)RO LT RTVS+240 (4.0/min)LC RF LF73 (6.1/min)LO LC LT342 (4.2/min)LF LC RF87 (7.3/min)LO LT RT Cat 3 11F3.3–161 (6.1/min)LC LF LT119 (9.9/min)LO RO LF241 (4.1/min)RF LT LF68 (5.7/min)LO RO LT330 (3.0/min)LF RF LT74 (6.2/min)LO RO LF Cat 4 11M4.5<1/year128 (2.8/min)LT LF RT127 (10.6/min)LO RO LTVS+224 (2.4/min)LF LC LT114 (9.5/min)LO RO LF330 (3.0/min)LC RT LT112 (9.3/min)LO RF RO Cat 5 1F3.2–150 (5.0/min)LF LT RC102 (8.5/min)LO LT RF239 (3.9/min)LC LT LF63 (5.3/min)LO LF LC344 (4.4/min)LC LF RF51 (4.3/min)LO RF LC Cat 6 2F2.1–164 (6.4/min)LF LC RT125 (10.4/min)LO RF RO223 (2.3/min)LC LF RF56 (4.7/min)LO RO RF337 (3.7/min)LC LF RF86 (7.2/min)LF LT LC Cat 7 2F2.8–139 (3.9/min)LC LF RO73 (6.1/min)LO RO RF254 (5.4/min)RF RT RC51 (4.3/min)LO LT LC342 (4.2/min)LF RC LC67 (5.6/min)RO LO RF Cat 8 1M3.4–127 (2.7/min)LC LF LO80 (6.7/min)LO RF RO226 (2.6/min)LC LF RC39 (3.3/min)LO RO LC313 (1.3/min)RF LC RC38 (3.2/min)RF LO RO Cat 9 12M5.3–148 (4.8/min)LF LC LT79 (6.6/min)RO LO LF237 (3.7/min)LC LF LO64 (5.3/min)LO RF RO350 (5.0/min)LC RC RT54 (4.5/min)LO RO LF Cat 10 1M4.4–120 (2.0/min)LC LF LT69 (5.8/min)LO RO LC227 (2.7/min)LC LF RO54 (4.5/min)RO LO RT327 (2.7/min)LC RF RT45 (3.8/min)LO LC ROIED counts are presented as the total number of IEDs, with values in parentheses indicating IED frequency (IEDs per minute). The top 3 regions indicate the 3 regions with the highest IED counts in each session. Seizure frequency is expressed as the number of seizures per year. –: no history of spontaneous seizures. VS+ indicates that vestibular stimulation was applied during the experiment. Abbreviations: EEG = electroencephalography; LT = left temporal; RT = right temporal; LF = left frontal; RF = right frontal; LO = left occipital; RO = right occipital; LC = left central cortex; RC = right central cortex. Subject information. IED counts are presented as the total number of IEDs, with values in parentheses indicating IED frequency (IEDs per minute). The top 3 regions indicate the 3 regions with the highest IED counts in each session. Seizure frequency is expressed as the number of seizures per year. –: no history of spontaneous seizures. VS+ indicates that vestibular stimulation was applied during the experiment. Abbreviations: EEG = electroencephalography; LT = left temporal; RT = right temporal; LF = left frontal; RF = right frontal; LO = left occipital; RO = right occipital; LC = left central cortex; RC = right central cortex. Experimental procedures, analyses, and evaluation methods Our exploratory pilot study was designed to develop and evaluate a methodological workflow of EEG-fMRI for future veterinary clinical application. Each cat underwent 3 experimental sessions, each consisting of scalp EEG performed outside the MRI scanner using a digital EEG system with subcutaneous needle electrodes (out-of-scanner EEG), placement of MRI-compatible cup electrodes, and simultaneous EEG and BOLD fMRI acquisition inside the MRI gantry (in-scanner EEG; Figure 1). Out-of-scanner EEG was performed under dexmedetomidine sedation, whereas EEG-fMRI was conducted under immobilization with additional rocuronium and low-dose sevoflurane, followed by reversal with atipamezole and sugammadex. The IEDs were manually counted in a 10-minute analyzable segment of the 20-25-minute EEG recording, and the 3 most active regions were identified for each session. For in-scanner EEG, MRI-compatible electrodes adapted for use in cats were connected to an MRI-compatible amplifier, using the same electrode components and amplifier as MRI-compatible EEG systems used in humans. After artifact removal, IED onsets were marked in BrainVision Analyzer 2 for fMRI analysis. The fMRI data were preprocessed and analyzed using SPM12 and Stolzberg’s feline template.22 The IEDs were modeled as event-related regressors convolved with the canonical hemodynamic response function (HRF) in a general linear model (GLM). The BOLD activation maps were evaluated sequentially using descending statistical thresholds. Maps were first assessed at a family-wise error-corrected threshold of P < .05 and an uncorrected threshold of P < .001. When activation was not consistently detected at these thresholds, a more liberal exploratory threshold was used to evaluate BOLD activation patterns (Figure 2). Experimental procedures and immobilization protocol. (A, B) Setup of the out-of-scanner scalp EEG recording. (C–F) MRI-compatible cap electrodes for cats and their placement. (G) Representative T1-weighted image and (H) echo planar imaging (EPI) acquired in the present study. (I) Overview of simultaneous EEG–fMRI acquisition. The cat’s head was positioned within a 16-channel flexible RF coil, and the connecting cables were secured with sandbags. Oxygen saturation, electrocardiography, and blood pressure were monitored during imaging. Flowchart of the preprocessing and analysis procedures used for generating BOLD activation images in this study. (A) fMRI preprocessing. (B) Artifact removal process for in-scanner EEG. (C) Generation and evaluation of BOLD activation images. Abbreviations: BOLD = blood oxygenation level dependent; T1WI = T1-weighted imaging; TPM = tissue probability map; EPI = echo planar imaging; AAS = average artifact subtraction; IEDs = interictal epileptiform discharges; GLM = general linear model. Evaluation consisted of primary and secondary analyses (Figure 3). Primary analyses assessed IED-related BOLD activation within the predefined amygdalo-hippocampal region using 2 approaches: session-level region of interest (ROI) analysis with small-volume correction (SVC) and cohort-level group analysis using a 1-sample t-test across all 30 sessions. Secondary analyses evaluated within-cat reproducibility across 3 sessions, regional concordance between out-of-scanner EEG and EEG-fMRI, and whole-brain anatomical distribution of activated clusters. For secondary analyses, activated regions were categorized into predefined anatomical groups, and the top 3 activated regions were selected for each session as described in the Appendix. Overview of the evaluation methods used for the results in this study. Primary analyses consisted of small-volume correction (SVC) analysis within the predefined amygdalo-hippocampal ROI and group-level analysis across all sessions. Secondary analyses consisted of within-cat reproducibility analysis, comparison with out-of-scanner EEG, and whole-brain anatomical distribution analysis. The 95% CIs for the observed proportions were calculated using the Wilson method in R. Detailed methods are provided in the Appendix. Results Devices In the our study, EEG-fMRI (ie, simultaneous EEG recording and BOLD fMRI acquisition) was technically completed in all 30 sessions using the modified MRI-compatible EEG electrodes and amplifier. Immobilization In all cats, the immobilization protocol using dexmedetomidine, rocuronium, and low-dose sevoflurane allowed stable acquisition during in-scanner EEG and fMRI acquisition without apparent motion artifacts. Although some cats moved during electrode placement, additional sedative doses within the predefined range achieved sufficient immobilization during EEG-fMRI. The median immobilization time (from the first dexmedetomidine administration to completion of imaging) was 108 minutes (range, 82-124 minutes; Supplementary Table S1). Out-of-scanner EEG Scalp EEG recordings were obtained under dexmedetomidine alone, as previously reported.10 The IED counts and the top 3 regions identified based on these counts are presented in Table 1. In-scanner EEG The MRI-compatible EEG cup electrodes achieved impedances <10 kΩ in all cats, and no electrode displacement was observed during scanning. The median electrode placement time was 12 minutes (range, 5-26 minutes; Supplementary Table S1). Gradient artifacts on in-scanner EEG were effectively removed using average artifact subtraction (AAS), with no substantial residual artifacts interfering with visual EEG interpretation. The EEG signals and IEDs could be visually assessed using average reference (AV) and bipolar montages. During the 12-minute analysis window, the median number of IEDs incorporated into the fMRI analysis was 71 (6 IEDs/minute; range, 38-127). fMRI Preprocessing All fMRI datasets were successfully preprocessed using the same software and workflow as used in humans,23 allowing reliable visualization of BOLD responses mapped onto structural images. BOLD activation results The BOLD activation maps were first evaluated using more stringent statistical thresholds. No activation survived correction for multiple comparisons at a family-wise error-corrected P < .05. Activation was identified in 2/30 sessions at P < .001 (uncorrected) and in 19/30 sessions at P < .01 (uncorrected), both with a minimum cluster size of 3 voxels (k ≥ 3). At the exploratory threshold of P < .05 (uncorrected, k ≥ 3), BOLD activation was detected in all 30 sessions (Figure 4, Supplementary Table S2). Accordingly, subsequent analyses in our pilot study were performed using this exploratory threshold. Example of the individual EEG-fMRI results (cat 4). Brain activation with a peak maximum in the left ventral hippocampus was observed in the first session of cat 4 (A, B; arrowheads indicate hippocampal activation), showing concordance with frequent left temporal spikes on scalp out-of-scanner (C) and in-scanner (D) EEGs. Primary analyses Small-volume correction analysis At the exploratory threshold, activation within the predefined amygdalo-hippocampal ROI was detected in 16/30 sessions (53.3%; 95% CI, 35.4%-70.6%). At the individual level, 5/10 cats (50.0%; 95% CI, 23.7%-76.3%) showed amygdalo-hippocampal activation in at least 2 of 3 sessions. Group-level analysis No activation cluster survived correction for multiple comparisons at a family-wise error-corrected threshold of P < .05 in the group-level analysis. The largest activation cluster extended from the left parahippocampal gyrus to the hippocampal body (P < .01 uncorrected, k ≥ 3; peak P < .001 uncorrected; k = 17). The lowest P-value was located in the left marginal gyrus (P < .001 uncorrected, k = 7). When the amygdalo-hippocampal ROI was applied, a similar activation pattern was observed within the left parahippocampal–hippocampal region (Figure 5, Supplementary Table S3). BOLD activation maps from the group-level analysis. Group-level analysis based on all sessions (n = 30) from 10 cats. Regions identified at P < .01 (uncorrected, k ≥ 3) are shown. Further details identified at P < .05 (uncorrected, k ≥ 3) are provided in the Supplementary Table S3. The peak maximum was located in the left marginal gyrus of the parietal cortex (arrow), and the largest cluster was observed extending from the left parahippocampal gyrus to the left hippocampal body (arrowhead). Secondary analyses Secondary analyses, including within-cat reproducibility, regional concordance between out-of-scanner EEG and EEG-fMRI, and whole-brain anatomical distribution, are summarized in Supplementary Tables S4 and S5. Briefly, reproducible activation was observed in 3/10 cats (30%; 95% CI, 10.8%-60.3%). Regional concordance between out-of-scanner EEG and EEG-fMRI was observed in 9/20 analyzable sessions (45%; 95% CI, 25.8%-65.8%), and 2/10 cats (20%; 95% CI, 5.7%-51.0%) showed both regional concordance and reproducible activation in at least 2 of 3 EEG-fMRI sessions. In the exploratory whole-brain anatomical analysis, activations included the hippocampus, parahippocampal gyrus, amygdala, piriform lobe, and other anatomically related regions. Discussion To the best of our knowledge, ours is the first exploratory study to describe and evaluate EEG-fMRI using a familial epilepsy model in cats as an initial step toward future veterinary clinical application. Previous animal fMRI studies have been reported in dogs, cats, and non-human primates.24–29 However, animal EEG-fMRI investigations have been limited to experimental models in monkeys, sheep, mice, and rats, and were primarily designed for translational research relevant to human neuroscience and medicine.30 These models were not intended for veterinary clinical application, and therefore, their MRI coils, electrode systems, and immobilization or awake protocols may not reflect clinical conditions encountered in veterinary practice. In our study, EEG-fMRI was performed using a clinically realistic setup based on equipment available in routine veterinary practice and components of clinical EEG-fMRI systems used in humans. The MRI-compatible EEG electrodes used for in-scanner EEG were adapted for use in cats by modifying the electrode array placement and cable configuration, whereas the electrode components and MRI-compatible EEG amplifier were the same as those used in MRI-compatible EEG systems designed for use in humans. The MRI scanner and radiofrequency (RF) coil also were standard clinical devices. Using this setup, simultaneous EEG recording and fMRI acquisition could be technically completed in all sessions. Key methodological challenges addressed in this pilot study included adapting MRI-compatible EEG electrodes for the head size of cats, developing an immobilization protocol that allowed stable image acquisition and interpretable EEG recordings, selecting an fMRI analysis workflow applicable to cats, and assessing whether IED-related BOLD responses could be visualized at an exploratory threshold under these conditions. The electrode system consisted of 13 electrodes, including 11 recording electrodes, 1 ground electrode, and 1 reference electrode. Because head size in cats varies less markedly than in dogs, this electrode arrangement may be applicable to a broad range of cats. In contrast, for future studies in dogs, substantial variation in head size and shape among breeds will require adjustments in the number and placement of electrodes as well as cable length. The immobilization protocol used in our study allowed stable EEG acquisition without apparent movement during MRI scanning. Although no established consensus exists, medetomidine or dexmedetomidine is commonly used for scalp EEG recording in veterinary medicine.10,31 However, animals sedated with these agents may remain responsive to loud auditory stimuli during MRI scanning.32 In our preliminary testing using dexmedetomidine continuous rate infusion (CRI) alone, the cats were likely aroused by MRI acoustic noise, resulting in movement during scanning and necessitating termination of the acquisition. Therefore, we combined the muscle relaxant, rocuronium, with low-dose sevoflurane to achieve movement-free immobilization. Rocuronium requires endotracheal intubation but has been shown not to affect EEG or cerebral blood flow in dogs, supporting its use as part of an immobilization protocol for EEG-fMRI.33 Sevoflurane can prolong muscle relaxation but may induce EEG abnormalities at high concentrations.34,35 However, no obvious EEG changes were observed at the low doses used in our study. Although awake fMRI has been achieved in primates, rodents, and trained dogs, such approaches are not practical for clinical veterinary patients because of facility limitations, required training, and variability in temperament. Therefore, immobilization protocols similar to the present approach may provide a practical framework for clinical EEG-fMRI studies in dogs and cats. Electroencephalographic recording suitable for visual identification of IEDs and subsequent analysis was obtained in all cats. Careful clipping and degreasing, followed by electrode paste application, enabled impedance values ≤10 kΩ in every session. Although EEG-fMRI in humans typically targets ≤5 kΩ, achieving such low impedance is often difficult in animals because of thick skin and dense hair coats.36 Therefore, setting a practical impedance target of ≤10 kΩ is reasonable for veterinary applications. Although some breeds may not require clipping, the thickness of the undercoat and the presence of skin oils often necessitate hair removal to ensure reliable data acquisition. In cases evaluated with epilepsy surgery in mind, shaving is generally acceptable, but shaving performed solely for EEG–fMRI examinations may be cosmetically undesirable, making owner consent more difficult to obtain. Four main types of artifacts are known to occur during EEG acquisition inside the MRI scanner: gradient artifacts generated by magnetic field gradients,37 motion artifacts caused by movement and associated induced currents,38 ballistocardiographic artifacts arising from cardiac pulsation and blood flow,39 and environmental artifacts produced by scanner room noise and vibration.40 The AAS method used in our study is a standard artifact removal technique used in humans for EEG-fMRI and produced interpretable EEG signals in our dataset as well.37 Appropriate immobilization, short cable lengths, and securing cables with sandbags further minimized artifact contamination.41 Ballistocardiographic artifacts also were negligible, as confirmed by simultaneous evaluation of ECG and EEG. However, interpretation of artifact-corrected EEG remains challenging even in humans, including for experienced readers.13 Therefore, additional artifact removal approaches, such as independent component analysis,42 principal component analysis,43 canonical correlation analysis,44 and optimal basis sets,45 should be considered to further improve EEG quality in future studies. In our study, IED (spike) detection was performed visually. Although visual marking is also standard in humans,13 automated approaches using independent component analysis or deep learning have been increasingly reported, and future work should incorporate these methods.46–48 In our study, the median IED count during the 12-minute scan was 71 spikes. In humans, approximately 93 spikes over a 60-minute scan (1.6 spikes/minute; range, 10-1132) are typically required to obtain clear BOLD responses.49 Despite the higher spike rate observed in our cats, no BOLD activation survived higher statistical thresholds and correction for multiple comparisons. This result may have been a consequence of multiple factors, including limitations in spike extraction, partial volume effects, and other physiological constraints, as discussed later. In our study, all IEDs observed during in-scanner EEG were extracted and used for fMRI analysis. However, future comparisons with out-of-scanner EEG, especially when isolating specific spike types, will require consistent immobilization conditions across recordings. In our study, out-of-scanner EEG was recorded under dexmedetomidine sedation alone, whereas in-scanner EEG was recorded under a combination of dexmedetomidine, rocuronium, and low-dose sevoflurane. The fMRI acquisition under the selected parameters allowed analyzable activation maps to be generated. No noticeable artifacts from EEG electrodes or cables were observed, consistent with reports in humans indicating that such artifacts generally are negligible.50 Our study was designed for clinical application, and therefore we used the standard 3 T MRI system and coil commonly employed in veterinary practice. In contrast, fMRI studies in rodents and primates often use dedicated coils and higher field strengths to maximize signal-to-noise ratio (SNR), which represents a limitation of our clinically oriented model. However, small animal-specific coils, such as strap-type coils now offered by some manufacturers, may further improve SNR in future studies. The scan duration in our study was 12 minutes, shorter than the 30-60 minutes typically used in humans for EEG-fMRI.41 This duration reflects the standard single-session fMRI acquisition time for the MRI system used in our study. Although shorter scans decrease anesthetic and immobilization stress, extending acquisition time in clinical cases where scanning opportunities are limited would allow detection of more IEDs and potentially yield clearer activation. Given the stable immobilization achieved in our study, future studies may consider extending scan duration to 20-30 minutes using the same protocol. Image preprocessing and visualization were performed using a brain template for cats previously employed in several MRI/fMRI studies.22 Cats exhibit relatively little inter-individual variation in brain morphology; therefore, this template provides sufficient anatomical accuracy for fMRI analysis. In contrast, dogs show considerable breed- and size-related variation in brain structure, and even individuals within the same breed can differ substantially. Although several dog templates have been reported, no high-precision template that accommodates all breeds or skull types is currently available.51–53 Therefore, future EEG-fMRI studies in dogs will likely require the creation of individualized or breed-specific templates. Regarding spatial smoothing in the preprocessing, studies in humans commonly use a full width at half maximum (FWHM) of 8 mm, whereas veterinary fMRI studies typically apply 2-4 mm depending on voxel size.26,27,29 In our study, we used an FWHM of 4 mm to facilitate clearer visualization of cluster size, but an FWHM of 2 mm may be more appropriate for finer-grained analyses. The primary analyses provided cohort-level findings of IED-related BOLD activation within the predefined amygdalo-hippocampal region of FSECs. The SVC analysis detected activation within the predefined amygdalo-hippocampal ROI in 16/30 sessions, and group-level analysis identified the largest activation cluster extending from the left parahippocampal gyrus to the hippocampal body. These findings are consistent with previous reports of amygdalo-hippocampal involvement in FSECs based on scalp EEG, structural MRI, intracranial EEG, and histopathology.12,19–21 Therefore, our results support the use of EEG-fMRI as a hypothesis-generating method for exploring IED-related network activity in this familial feline model. In contrast, the secondary analyses showed limited within-cat reproducibility and regional concordance with out-of-scanner EEG. Reproducible BOLD activation was observed in 3/10 cats, and regional concordance between out-of-scanner EEG and EEG-fMRI was observed in 9/20 analyzable sessions. These values were lower than those reported in EEG-fMRI studies in humans: a previous study54 reported reproducible findings in 5/8 patients who underwent repeated EEG-fMRI examinations at 3 T, whereas another study55 reported concordant BOLD responses in 29/33 patients with active EEG and contributory responses in 21/33 patients. These findings indicate substantial inter-session variability and suggest that, under the present protocol, EEG-fMRI is not yet sufficient as a standalone method for individual-level EZ localization. Future studies should improve spike classification, harmonize immobilization conditions between EEG recordings, and focus on clinically confirmed drug-resistant epilepsy cases to better evaluate individual-level reliability. The FSECs have been reported as a familial feline model showing temporal lobe epilepsy-like features, including limbic seizure semiology and amygdalo-hippocampal abnormalities.12,19–21 Although the primary analyses suggested amygdalo-hippocampal involvement at the cohort level, some sessions showed BOLD activation outside the amygdalo-hippocampal region or discordance between scalp EEG and EEG-fMRI. In humans with epilepsy, EEG-fMRI BOLD responses can extend beyond the presumed EZ and may represent propagation pathways or broader epileptic networks, rather than only the primary epileptogenic region.13,56,57 Therefore, such discordant or extra-amygdalo-hippocampal activations do not necessarily indicate simple false-positive findings, but also may reflect spatially distributed epileptic network activity that cannot be captured by scalp EEG alone. However, because intracranial EEG or surgical validation was not performed in our study, these possibilities remain speculative, and the observed activations cannot be interpreted as definitive localization of the EZ. Future studies should evaluate whether EEG-fMRI activation corresponds to intracranial EEG findings or surgical outcomes in cats with drug-resistant epilepsy undergoing epilepsy surgery, as reported in studies of humans.58,59 A major limitation of our study was the use of an uncorrected statistical threshold for BOLD activation. No activation survived correction for multiple comparisons. Therefore, our findings should be interpreted as exploratory and hypothesis-generating. This lack of corrected significance likely reflects both limited statistical power and decreased BOLD sensitivity under the present protocol. Potential contributing factors include small sample size, short acquisition time, use of clinical rather than dedicated small animal MRI hardware, and sedation- or immobilization-related attenuation of neurovascular coupling and vascular reactivity. Additional factors may include spatial resolution- and analysis-related limitations arising from the use of clinical MRI acquisition settings in small feline brains, including relatively large voxel dimensions compared with feline brain structures and increased partial-volume effects. Other possible contributors include suboptimal acquisition and preprocessing parameters for veterinary EEG-fMRI, potential mismatch between the canonical HRF derived from humans and the BOLD response of cats, and limited physiological noise correction. Future studies should optimize acquisition and analysis parameters, including scan duration, receiver coils, spatial smoothing, physiological noise correction, species-specific hemodynamic modeling, and the establishment of statistical thresholds appropriate for veterinary EEG-fMRI. Validation in clinically confirmed epilepsy surgery candidates using intracranial EEG, multimodal presurgical evaluations, and surgical outcomes also will be necessary. Our study had some additional limitations related to its exploratory methodological design. First, FSECs were selected as a biologically relevant model for establishing EEG-fMRI methodology, but only a small number of cats were included, and several did not have documented clinical seizures. Therefore, the present cohort was not suitable for evaluating individual-level diagnostic accuracy. Second, the presumed amygdalo-hippocampal involvement was based on previous studies of this familial strain, but intracranial EEG or surgical confirmation was not performed in the cats included in our study. In addition, because the MRI-compatible cup electrodes used in our study were adapted from commercially available EEG electrodes designed for use in humans rather than newly developed specifically for cats, we did not perform additional technical validation, such as impedance comparisons with SC needle electrodes or power spectral density comparisons between electrode types. Third, out-of-scanner EEG and in-scanner EEG were recorded under different immobilization conditions, which may have affected spike characteristics and regional concordance. These limitations reflect the initial methodological nature of our study, and further evaluation in clinically confirmed drug-resistant epilepsy cases undergoing presurgical assessment is the next necessary step. Validation against intracranial EEG, multimodal imaging, and postoperative outcomes will be required to determine whether EEG-fMRI can contribute to clinical decision-making in veterinary epilepsy surgery. Once optimized and validated, EEG-fMRI also may provide a noninvasive framework for investigating epileptic networks, drug-related changes in brain activity, sleep-related EEG phenomena, and event-related brain responses in companion animals.60–62 Conclusion and clinical importance Our exploratory pilot study showed that simultaneous EEG-fMRI can be technically implemented in FSECs using a clinically realistic veterinary MRI setup. Cohort-level amygdalo-hippocampal activation was observed at an exploratory threshold, but individual-level reproducibility, EEG concordance, and statistical robustness remained limited. Further validation in clinical epilepsy surgery candidates, including comparison with intracranial EEG and surgical outcomes, is required before EEG-fMRI can be applied to clinical EZ localization. Supplementary Material