Journal of neuroimaging : official journal of the American Society of Neuroimaging
May 1, 2026 · via Europe PMC
Impact of Synthetic Lesional MR Images in Automated Focal Cortical Dysplasia Detection in Low-Data Scenarios.
Kaur P, Ouaalam H, Kandemirli S, Prabhu SP, Warfield SK.
AI summary — for doctors, generated from the abstract only
Synthetic FCD MRIs from a conditional generative network were hard for experts to distinguish from real scans and modestly improved automated lesion detection in low-data settings.
What was studied
Retrospective multi-site study using T1w and FLAIR MRI from 131 FCD patients and 90 controls to generate synthetic lesional images and train nnU-Net detectors with real, real-plus-synthetic, or expanded real data.
Key findings
- Experts identified real vs synthetic with only 60% (T1w) and 70% (FLAIR) accuracy
- Synthetic augmentation raised sensitivity by 8.14% (p = 0.12, not significant) and improved lesion-site confidence (0.83 to 0.89, p = 0.02)
- Expanded real-data model reached 73.8% sensitivity; specific statistic not fully captured in the abstract
- Authors estimate about 20% reduction in labeled data needs
Clinical relevance
This suggests synthetic data may ease annotation burden for FCD detection tools, though real data remains superior when available and the method is not a clinical tool itself.
Who should care
Limitations
Retrospective study; sensitivity gain from synthetic data was not statistically significant.
Synthetic MRIs can partly substitute for scarce labeled FCD data, but real data still wins.
This is an AI-generated summary of the published abstract, not a clinical recommendation. It may not reflect the full text or subsequent updates — verify against the original source before acting on it.
Abstract
Background and purpose
Automated detection of focal cortical dysplasia (FCD) requires large volumes of voxelwise-lesion-delineated MRI data, which are difficult to acquire. This study aims to generate synthetic MRI data exhibiting FCD, assess its realism, and evaluate its impact on automated FCD detection-particularly in reducing the need for manual annotations.
Methods
T1-weighted (T1w) and T2-weighted-fluid-attenuated inversion recovery (FLAIR) MRI scans from 131 FCD patients and 90 healthy controls from multiple (3) sites were retrospectively studied. Synthetic MRIs were generated by conditioning a generative network on binary FCD mask. Two neuroradiologists identified real images from a random set of 14 real and 14 synthetic scans. Three nnU-Net models were trained to detect FCD using (i) real-only (35-FCD/35-controls), (ii) real (35-FCD/35-controls) + synthetic augmentation, and (iii) expanded real data (70-FCD/70 controls).
Results
Experts showed limited ability to distinguish real from synthetic images, with classification accuracy of 60% for T1w and 70% for FLAIR (inter-rater agreement κ = 0.86). Augmenting automated FCD detection with synthetic data increased sensitivity by 8.14% (p = 0.12) and improved model confidence at true lesion sites (0.83 ± 0.11 to 0.89 ± 0.12; p = 0.02). The expanded real-data model further improved sensitivity to 73.8% (p < 0.001) and confidence to 0.90 ± 0.14 (p = 0.01).
Conclusion
Conditional generative networks can generate realistic synthetic FCD-MRIs, reducing labeled data needs by ∼20% while maintaining equivalent sensitivity. Equivalent amounts of real data, when available, remain more effective than synthetic augmentation.
DOI
10.1111/jon.70137
PMID
42240056
Research metadata (titles, abstracts, authors, publication dates) is sourced from Europe PMC, which aggregates content from PubMed/MEDLINE, PubMed Central, and other providers under the U.S. National Library of Medicine (NLM/NCBI). NLM does not claim copyright on abstracts; the original journal publishers or authors may hold rights to the full articles. AI-generated summaries are derived only from the abstract, never the full text, and are always labeled as AI-generated.
Want more like this?
Browse more from the feed. A free account unlocks the full feed, saving and communities.
Back to the research feed