External validation of automated focal cortical dysplasia detection using morphometric analysis

Area of Science:

• Neurology
• Medical Imaging
• Artificial Intelligence

Background:

• Focal cortical dysplasias (FCDs) are a primary cause of drug-resistant focal epilepsy.
• Conventional MRI often fails to detect FCDs, complicating diagnosis and treatment planning.
• Morphometric analysis of T1-weighted MRI, like MAP18, aids visual detection but requires expertise.

Purpose of the Study:

• To develop and validate an artificial neural network (ANN) classifier for automated FCD detection.
• To assess the ANN's performance and generalizability using morphometric maps from MAP18.
• To create a robust, clinically viable tool for presurgical epilepsy evaluation.

Main Methods:

• A feed-forward ANN was trained and cross-validated on 113 FCD patients and 362 controls.
• The ANN utilized morphometric output maps generated by the Morphometric Analysis Program (MAP18).
• Performance was validated on an independent dataset of 60 FCD patients and 70 controls, using data from 13 different scanners.

Main Results:

• The ANN achieved 87.4% sensitivity and 85.4% specificity during cross-validation.
• On the independent dataset, the ANN demonstrated 81.0% sensitivity and 84.3% specificity.
• The method showed robust performance, largely independent of scanner or MR-sequence parameters.

Conclusions:

• The developed ANN provides robust automated detection of FCDs.
• The method exhibits strong generalizability across different imaging sites and parameters.
• This AI-driven approach offers a clinically viable tool for presurgical diagnosis of drug-resistant focal epilepsy.