In a retrospective study, synthetic interictal SPECT images generated using a deep learning model produced subtraction ictal-interictal SPECT co-registered to MRI maps that closely matched those obtained through conventional real imaging. The experimental framework, termed SySCOM, preserved the spatial localisation of brain regions exhibiting increased perfusion during seizures. Blinded expert readers evaluating the outputs demonstrated high levels of diagnostic agreement when interpreting synthetic and real images to identify epileptogenic foci.
Presurgical localisation of epileptogenic regions relies on interictal SPECT as a critical component of standard SISCOM analysis. In routine clinical practice, however, the diagnostic reliability of acquired interictal SPECT scans can be compromised by uncertainty regarding whether a scan represents a genuine seizure-free baseline state. Moreover, most patients undergoing presurgical evaluation must also complete an interictal fluorodeoxyglucose PET scan and an MRI scan, making methods that could reduce the overall number of required diagnostic procedures highly desirable.
To investigate whether a machine learning-generated scan could provide a dependable alternative baseline to acquired interictal SPECT, researchers developed a framework based on a three-dimensional generative adversarial network. The neural network was trained on PET scans acquired from 40 patients diagnosed with refractory focal epilepsy to synthesise realistic interictal SPECT image volumes.
The synthesis performance of the deep learning architecture was evaluated on an independent validation cohort of nine previously unseen patients. Quantitative comparisons between the generated and acquired scans were conducted using standard image quality metrics, specifically the structural similarity index measure, peak signal-to-noise ratio, and root-mean-square error. Both synthetic and acquired interictal images were subsequently processed into SISCOM and SySCOM outputs following standard clinical imaging protocols.
To determine clinical concordance, five clinicians experienced in interpreting SPECT images in epilepsy performed blinded visual localisation of hyperperfusion across 18 predefined anatomical brain regions. The research team evaluated observer-based similarity, diagnostic agreement, and reading reproducibility between standard SISCOM and synthetic SySCOM outputs.
The synthetic interictal SPECT images showed high quantitative similarity to real scans, achieving a mean structural similarity index measure of 0.94, a mean peak signal-to-noise ratio of 27.9 decibels, and a mean root-mean-square error of 0.041. In clinical visual localisation tasks, SISCOM and SySCOM demonstrated good agreement, marked by a Dice score of 70 percent and an overall agreement rate of 89 percent across all evaluated subjects and anatomical regions. Individual reader agreement across sessions remained high, reflecting stable clinical interpretation between real and synthetic image modalities.
The study authors concluded that machine learning-synthesised interictal SPECT can produce SySCOM maps that are visually and quantitatively comparable to standard SISCOM. By generating baseline data computationally, the technique directly addresses the uncertainty surrounding whether an acquired interictal SPECT scan captures a true seizure-free state. The researchers noted that while SySCOM represents a feasible clinical approach, it requires further clinical validation before routine implementation in patient care.
Subject of Research: Synthetic interictal SPECT imaging for epileptogenic focus localisation
Article Title: SySCOM: epileptogenic focus localisation using synthetic interictal SPECT images in SISCOM
Article References: Fard, A. S., Reutens, D. C., Ramsay, S. C., Phamnguyen, J., Fong, W., Bhatt, M. K., Goodman, S. J., & Vegh, V. (2026). SySCOM: epileptogenic focus localisation using synthetic interictal SPECT images in SISCOM. EJNMMI Physics. https://doi.org/10.1186/s40658-026-00928-7
Image Credits: AI Generated
DOI: 10.1186/s40658-026-00928-7
Keywords: epilepsy, refractory focal epilepsy, SPECT, SISCOM, SySCOM, deep learning, generative adversarial network, presurgical evaluation, epileptogenic, focus, localisation, synthetic
News Source: Denise Maddox. (November 9, 2026). Research report: SySCOM: epileptogenic focus localisation using synthetic interictal SPECT images in SISCOM. Scienmag.



