Automated Whole-Brain Focal Cortical Dysplasia Detection Using MR Fingerprinting With Deep Learning.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Epilepsy Research

Background:

• Focal cortical dysplasia (FCD) is a leading cause of drug-resistant focal epilepsy.
• Detecting FCD on standard clinical MRI remains a significant challenge for clinicians.
• Magnetic resonance fingerprinting (MRF) offers rapid and precise quantitative tissue property measurements.

Purpose of the Study:

• To develop and evaluate a whole-brain deep learning (DL) framework utilizing MRF for FCD detection.
• To assess the efficacy of MRF-derived quantitative imaging features for identifying FCD lesions.
• To compare the performance of the MRF-DL model against conventional MRI and existing analysis pipelines.

Main Methods:

• A cohort of patients with pathologically confirmed FCD and healthy controls (HCs) underwent 3D whole-brain MRF and clinical MRI.
• MRF data were processed to generate T1, T2, gray matter (GM), and white matter (WM) tissue fraction maps.
• A U-Net deep learning model was trained using various combinations of MRF-derived maps and evaluated via cross-validation.

Main Results:

• The optimal MRF-DL model incorporated T1w, T1z, T2z maps, tissue fraction maps, and morphometric maps, achieving 80% patient-level sensitivity with 1.7 false positives per patient.
• The MRF-DL model demonstrated consistent sensitivity across FCD subtypes and locations, outperforming models using only clinical MRI data.
• The proposed MRF-DL framework surpassed the established MAP18 pipeline in sensitivity, false positive rate, and lesion overlap.

Conclusions:

• The developed MRF-DL framework is effective for whole-brain FCD detection.
• Multiparametric quantitative features derived from a single MRF scan provide valuable input for DL-based detection of subtle epileptic lesions.
• This approach holds promise for improving the diagnosis and management of FCD in epilepsy patients.