Localizing the epileptogenic zone using deep learning and neuroimaging: A systematic review.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Epilepsy Research

Background:

• Deep learning (DL) offers potential for identifying the epileptogenic zone (EZ) to improve surgical outcomes in drug-resistant epilepsy.
• Systematic review of DL applications in EZ localization from neuroimaging data.

Purpose of the Study:

• Synthesize evidence on DL-assisted EZ localization.
• Identify methodological trends, limitations, and future directions.
• Bridge the gap between clinical translation and technological advancements.

Main Methods:

• Systematic literature search of PubMed, Scopus, and Embase.
• Assessment of study bias and applicability using PROBAST+AI tool.
• Extraction of methodological details and performance metrics.

Main Results:

• Thirty-six studies reviewed, primarily segmenting epileptogenic lesions with structural MRI.
• Focal cortical dysplasia was the most common target; fully convolutional networks were the predominant DL architecture.
• Two-thirds of studies had high risk of bias and limited clinical applicability; progress in fine-grained localization was moderate.

Conclusions:

• Methodological limitations impede clinical translation of DL for EZ localization.
• Recommendations provided to address identified limitations.
• Future work should focus on standardized evaluation, uncertainty quantification, foundation models, and synthetic data.