AI in epilepsy neuroimaging

Comprehensive Summary

This paper examines how artificial intelligence is being applied to brain imaging in epilepsy, covering a broad range of uses from detecting lesions to improving image quality. The researchers summarized existing published studies across multiple AI applications in epilepsy neuroimaging, evaluating different machine learning approaches, the datasets used to train them, and how well each tool performed and generalized to new patients. The most developed application is the detection of focal cortical dysplasias, which are small abnormal patches of brain tissue that cause drug-resistant seizures and are missed by expert radiologists in nearly one third of cases, where the best-performing AI tool detected 64% of previously missed lesions with very few false positives. AI has also shown strong results in identifying and lateralizing hippocampal sclerosis, the most common cause of temporal lobe epilepsy, with one model correctly identifying the affected side in 97% of patients including those with normal-appearing MRIs. Additional applications include predicting seizure risk after brain tumors or traumatic injury, locating where seizures start in the brain, and even generating synthetic imaging data to reduce radiation exposure or improve low-quality scans from resource-limited settings. The authors emphasize that while these tools are promising, most remain research-grade and have not yet been formally regulated or integrated into routine clinical workflows.

Outcomes and Implications

This work is important because epilepsy affects roughly 1 in 100 people, and missed or delayed diagnosis of treatable structural causes directly delays access to potentially curative surgery, which remains underutilized globally. Several of the described AI tools are already openly available and have been validated across multiple hospitals, meaning they are approaching the point of genuine clinical usefulness for neurologists, neuroradiologists, and epilepsy surgeons. The authors are cautiously optimistic about near-term clinical translation, noting that the next steps including regulatory approval as medical devices, integration into hospital imaging systems, and ensuring clinicians can understand and trust AI outputs are the remaining barriers before these tools routinely reach patients.