For this retrospective study, the database of the Department of Nuclear Medicine of the University Medical Center Hamburg-Eppendorf was searched for patients who had undergone both ictal and interictal brain perfusion SPECT for presurgical evaluation of epilepsy. The following inclusion criteria were applied: (I1) the same tracer, either 99mTc-HMPAO or 99mTc-ECD, was used for ictal and interictal SPECT, (I2) the same double-head camera was used for ictal and interictal SPECT, (I3) both SPECT images and structural MRI were digitally available for consistent retrospective image processing, (I4) scalp video-EEG recording during the ictal injection was available for retrospective analysis and clearly identified a seizure, (I5) the latency of ictal tracer injection after electrical seizure onset was ≤ 60 s, (I6) at the interictal tracer injection, the patient was seizure-free since at least 24 h, (I7) the time interval between both SPECT acquisitions was ≥ 48 h (in order to avoid contamination from residual radioactivity), and (I8) the time interval between ictal and interictal SPECT was ≤ 7 days (at our site, both scans are usually performed during the same inpatient stay lasting a maximum of 7 days, Monday–Monday). Severe image artifacts or technical problems hampering data interpretation (e.g., head motion during the ictal SPECT acquisition) led to exclusion. There were no eligibility criteria with respect to the suspected localization of the SOZ, MRI findings or prior therapy. This was to guarantee that the included patient sample was representative of brain perfusion SPECT for presurgical evaluation of epilepsy patients in clinical routine at our site (tertiary referral epilepsy specialist center).

The eligibility criteria led to the inclusion of 177 patients. Demographic and clinical data are summarized in Table 1. The majority of the patients had previously been included in a study on the impact of the tracer, 99mTc-HMPAO versus 99mTc-ECD [13], and/or in a study on the impact of the post-injection seizure duration on ictal brain perfusion SPECT [14]. A small subset of the patients had previously been included in a study on covariance pattern analysis of ictal brain perfusion SPECT for predicting the outcome of epilepsy surgery [15].

Information on the seizure outcome at 12 months after epilepsy surgery was available in 44 patients. The seizure outcome was favorable (Engel outcome scale I or II [16, 17]) in 33 of these patients (75.0%). Among the 33 patients with favorable 12 months seizure outcome, surgery had been on the temporal lobe in 31 patients (93.9%, Engel I-A/B/C/D: n = 19/6/1/0, Engel II-A/B/C/D: n = 2/1/0/2). Surgery was on the frontal lobe in the remaining 2 patients (6.1%).

Ictal and interictal injections of 99mTc-ECD (n = 36) or (stabilized) 99mTc-HMPAO (n = 141) were performed in an inpatient setting under scalp video-EEG monitoring in the Department of Neurology and Epileptology of the Protestant Hospital Alsterdorf. The median time interval between ictal and interictal SPECT was 2 days (interquartile range 2–4 days). Ictal SPECT was performed first in the vast majority of the patients (98.8%).

For SPECT imaging, patients were transported from the Protestant Hospital Alsterdorf to the University Medical Center Hamburg–Eppendorf (about 20 min drive). SPECT projection data were acquired for 40 min net acquisition duration with a double-head camera (either Siemens Symbia T2 or Siemens E.CAM) equipped with fan-beam or low-energy high-resolution collimators and angular steps of either 2.8° or 3.0°.

The projection data were reconstructed into transaxial SPECT images with a voxel size of 3.9 × 3.9 × 3.9 mm3 using filtered backprojection with a Butterworth filter of order 5 and cutoff 0.6 cycles/pixel (= 1.5 cycles/cm). The reconstructed images were postfiltered with an isotropic Gaussian kernel with 8 mm full-width-at-half-maximum. Uniform post-reconstruction attenuation correction was performed according to Chang (μ = 0.12/cm), scatter correction was not applied.

The ictal and the interictal SPECT image of a given patient were co-registered, first to each other and then to the patient’s MRI. The Coregister tool of the Statistical Parametric Mapping software package (version SPM12) was used for this purpose. Then, both SPECT images were spatially normalized (affine) to the anatomical space of the Montreal Neurological Institute (MNI) using the Normalize tool of SPM12. The interictal SPECT was used as the object image for spatial normalization. Depending on the tracer in this patient, a custom 99mTc-HMPAO template or a custom 99mTc-ECD template in MNI space was used as the target for spatial normalization [13].

In preparation of voxel-based statistical testing for ictal hyperperfusion and interictal hypoperfusion, spatially normalized SPECT images were smoothed by convolution with an isotropic 3-dimensional Gaussian kernel with 15 mm full-width-at-half-maximum and then scaled to the individual mean tracer uptake in a standard cerebrum parenchyma mask predefined in MNI space. The resulting images were transformed to z-score maps relative to the voxel-wise mean and the voxel-wise standard deviation of spatially normalized, smoothed, and scaled tracer uptake in a custom age-matched 99mTc-HMPAO normal database or a custom age-matched 99mTc-ECD normal database, depending on the tracer in the individual patient (Supplementary Fig. 1) [11, 15]. The SPECT images in the normal databases had been acquired with the same cameras using the same acquisition and reconstruction protocol as the SPECT images included in the current analyses. The cutoff z ≥ 3.0 on the z-score was used to identify significant ictal hyperperfusion and significant interictal hypoperfusion [18]. This conservative cutoff is used in clinical routine at our site to limit the rate of false-positive clusters.

SISCOM analysis was performed using a custom approach described previously [19]. In brief, the relative difference between ictal and interictal SPECT was computed as

$$} = \left( }*}}} \right) \, ./}$$

(1)

where “sictal” (“sinterictal”) is the 15 mm-smoothed ictal (interictal) SPECT image in MNI space, “sf” is a scale factor, and “./” denotes voxelwise division of the 3-dimensiuonal image matrices. The computation of relDiff was restricted to the standard cerebrum parenchyma mask in MNI space. The scale factor sf was determined by minimizing the sum of the squared relDiff values across all voxels in the parenchyma mask:

$$\sum\nolimits_}\;}\;}}} }_}}^ } \to }.$$

(2)

The resulting relDiff image was transformed to a z-score map by scaling the relDiff map to the standard deviation (SD) of relDiff across all voxels in the parenchyma mask:

For the next iteration, the scaling mask was restricted to voxels in the parenchyma mask with

$$}\left( }} \right) < ,$$

(4)

in order to reduce the impact of ictal hyperperfusion and interictal hypoperfusion on scaling and z-score transformation. Steps (2)–(4) were repeated for up to 30 iterations. The iteration was stopped earlier if the volume of the scaling mask dropped below 300 ml or when the change in volume of the scaling mask dropped below 10 ml. The final z-relDiff map was transformed to the individual patient space using the inverse of the affine transformation from patient to MNI space. Trilinear interpolation was used to write the transformed z-redDiff map to the matrix grid in patient space. The resulting map was overlaid to the patient’s MRI. A cutoff z ≥ 2.0 on z-relDiff was used to identify significant effects in the SISCOM image [11, 20].

The SPECT data were interpreted by three independent readers (KS, AK, RB) in three different settings using a standardized display for each setting: “interictal only” (Fig. 1), “ictal only” (Fig. 2) and “full” setting. The standard display for the “full” setting consisted of an 8-page pdf document starting with the two pages from the “ictal only” display (Fig. 2) followed by the two pages from the “interictal only” display (Fig. 1). The next two pages combined ictal and interictal uptake images in MNI space on one page, and the statistical maps of ictal hyper- and interictal hypoperfusion on another page (Supplementary Fig. 2). These pages aimed to simplify the side-by-side interpretation of ictal and interictal findings (no scrolling through different pages required). The last two pages displayed the full (non-thresholded) z-relDiff map and the conventional SISCOM map, that is, the z-relDiff thresholded at z ≥ 2.0 overlaid to the individual's MRI (Fig. 3).

Standardized display for visual reading in the “interictal only” setting. The upper and lower parts were provided as page 1 and page 2 of a 2-page pdf-document. On the first page, the left side shows the patient’s interictal perfusion SPECT coregistered to the individual T1-weighted MRI shown on the right. On the second page, the left side shows the patient’s interictal perfusion SPECT after spatial normalization to MNI space, the right side shows the statistical hypoperfusion map thresholded at z ≥ 3.0 and overlaid to the patient’s interictal SPECT. The upper threshold of the color bar for the display of the SPECT images was set to the maximum of the 3-dimensional SPECT image volume, separately for each SPECT scan. The lower threshold of the color bar was set to zero in all cases. The example images were acquired with 99mTc-HMPAO in a 28y old woman with normal MRI

Standardized display for visual reading in the “ictal only” setting. The upper and lower parts were provided as page 1 and page 2 of a 2-page pdf-document. On the first page, the left side shows the patient’s ictal perfusion SPECT coregistered to the individual T1-weighted MRI shown on the right. On the second page, the left side shows the patient’s ictal perfusion SPECT after spatial normalization to MNI space, the right side shows the statistical hyperperfusion map thresholded at z ≥ 3.0 and overlaid to the patient’s ictal SPECT. The example images are from the same patient as in Fig. 1. Ictal SPECT was performed with 700 MBq 99mTc-HMPAO injected 36 s after electrical seizure onset. The seizure continued for 70 s after the tracer injection

Standardized display for visual reading of SISCOM results in the “full” setting. The upper and lower parts were provided as page 7 and page 8 of an 8-page pdf-document. On page 7, the left side shows the full (non-thresholded) z-relDiff map coregistered to the individual T1-weighted MRI shown on the right. The z-relDiff map is displayed with a split color table to support the discrimination between positive (ictal > interictal) and negative (interictal > ictal) z-scores. On page 8, the left side shows the SISCOM map thresholded at z ≥ 2.0, the right side again shows the individual T1-weighted MRI. The example images are from the same patient as in Figs. 1, 2

Thus, three different pdf documents were created for each patient, resulting in a total of 531 different documents (= 177 patients * 3 settings). These documents were presented to the readers in a fully randomized order (with respect to both patient and setting) during a single reading session spread over several days. The readers were blinded to all clinical information except sex, age, and tracer, in order to avoid bias from clinical information. In particular, the readers were blinded to the suspected localization of the SOZ following standard diagnostic work-up prior to SPECT.

The readers were asked to first lateralize the SOZ: right hemisphere, left hemisphere, or no evidence of the SOZ. If a reader lateralized the SOZ to the right or to the left hemisphere, she/he first rated her/his confidence regarding the lateralization according to a 5 point scale (1 = very low confidence,…, 5 = very high confidence) and then localized the SOZ within a brain lobe (temporal, frontal, parietal, occipital) in the selected hemisphere. The category “no evidence of SOZ” was restricted to cases with normal tracer uptake. In cases with ≥ 2 SOZ candidates considered equally likely to represent the epileptogenic zone, the readers were asked to select one SOZ candidate and, if applicable, express their uncertainty by a low confidence score.

One of the readers had about 20 years of previous experience in reading brain perfusion SPECT in the presurgical evaluation of epilepsy patients. The other two readers did not have experience in reading brain perfusion SPECT before this study. In order to account for this, the readers were asked to adhere to the detailed recommendations for visual interpretation given in section “Criteria for visual identification of the seizure onset zone” in the supplementary material. In addition, the readers were provided with a figure presenting the voxel-wise mean and the voxel-wise coefficient of variance in the custom 99mTc-ECD and 99mTc-HMPAO normal databases as reference for the normal distribution pattern of these tracers in the brain (Supplementary Fig. 1). Before reading the 531 pdf documents for the current study, each reader underwent a training session with 12 randomly selected cases. The results of this training session were discussed among the readers to reduce between-readers variability.

Proportions are given as percentages, continuous variables are characterized by their median and interquartile range.

Between-readers agreement of the lateralization of the SOZ (right hemisphere, left hemisphere, no SOZ candidate) was assessed by Fleiss’ kappa (κ), separately for each setting. The statistical significance of pairwise κ differences between settings was assessed by checking the 83.4% confidence intervals (CI) for the corresponding κ estimates for overlap (non-overlapping 83.4% CIs indicate statistical significance with a 5% type 1 error probability) [21]. The 83.4% CI was computed as κ ± 1.385 * standard error [22].

To test for reader-independent between-setting effects on the lateralization of the SOZ, the lateralization scores and the lateralization confidence scores of the three readers were combined into an overall “lateralizing” 3-score: “non-lateralizing”, “lateralizing with low confidence”, and “lateralizing with high confidence”. A case was considered "lateralizing with high confidence” if the three readers agreed on the lateralization of the SOZ (either right or left hemisphere) and each reader scored her/his confidence regarding the lateralization with ≥ 4 of 5 points on the confidence scale. A case was considered "lateralizing with low confidence” if the three readers agreed on the lateralization of the SOZ (either right or left hemisphere) and at least one reader rated her/his lateralization confidence with less than 4 points. All other cases were considered “non-lateralizing” (including cases scored as “no SOZ candidate” by at least one reader and cases with disagreement among the readers regarding the hemisphere). The impact of the setting on the proportion of “non-lateralizing”, “lateralizing with low confidence”, and “lateralizing with high confidence” cases was tested by repeated measures analysis of variance (ANOVA) with the lateralizing 3-score as dependent variable and setting (“interictal only”, “ictal only”, “full”) as within-subjects factor. Post-hoc testing of pairwise differences between two settings was performed with repeated measures ANOVA restricted to those two settings. Patient-level between-settings differences of the lateralizing 3-score were assessed by 3 × 3 cross tables.

A case was considered “localizing” if it was lateralizing (with low or high confidence) and the three readers also agreed on the same lobe within the same hemisphere. The primary aim of the localization was to characterize the patient sample with respect to temporal versus extratemporal SOZ. The localization analysis therefore was restricted to the “full” setting.

Finally, SOZ lateralization was correlated with the operated hemisphere in the 31 patients with favorable seizure outcome 12 months after temporal lobe epilepsy surgergy, separately for each setting.

The statistical analysis was performed with SPSS (version 29). An effect was considered statistically significant if two-sided p < 0.05.