Can deep learning detect focal cortical dysplasia in pediatric drug-resistant epilepsy?
Focal cortical dysplasia (FCD) is a common cause of drug-resistant epilepsy in children, but it can be hard to see on standard MRI scans. Deep learning, a type of artificial intelligence, is being tested to help radiologists find these subtle brain lesions. Studies show that deep learning models can detect FCD with reasonable accuracy, though they are not perfect and still need expert review.
What the research says
A 2025 study tested two deep learning models, MELD Graph and 3D-nnUNet, on MRI scans from 71 children with drug-resistant epilepsy 2. MELD Graph found FCD with a precision of 0.85 and recall of 0.52, while 3D-nnUNet had a precision of 0.91 and recall of 0.48 2. In children whose initial MRI reports were negative (no visible lesion), MELD Graph produced more false positives (0.53 per patient) than 3D-nnUNet (0.14 per patient) 2. At the patient level, MELD Graph had slightly higher sensitivity (0.63 vs. 0.54) 2.
A 2025 systematic review and meta-analysis of 41 studies found that AI models for FCD detection had a pooled sensitivity of 0.81 and specificity of 0.92 in internal tests, but performance dropped in external validation (sensitivity 0.73, specificity 0.66) 6. This suggests that while deep learning is promising, its accuracy varies across different hospitals and scanners.
Another 2025 study used a different deep learning method with MR fingerprinting, a new MRI technique, and found it could detect FCD across the whole brain 5. The study reported that combining multiple MRI maps (T1, T2, gray matter, white matter) improved detection 5. A 2024 review also noted that advanced imaging and post-processing techniques, including deep learning, can help identify the epileptogenic zone in drug-resistant epilepsy, especially when standard MRI is negative 7.
What to ask your doctor
- Could a deep learning analysis of my child's MRI help find a subtle FCD that was missed on the initial report?
- What are the false positive rates for these AI tools, and how would false positives be handled?
- Is the deep learning model used at your center validated on pediatric patients?
- Would my child need additional imaging, like MR fingerprinting, to improve detection?
- How do you combine AI findings with other tests like EEG or clinical symptoms to decide on surgery?
This question is drawn from common patient questions about this topic and answered using cited medical research. We do not provide individualized advice.