Convolutional Neural Networks for Pediatric Refractory Epilepsy Classification Using Resting-State Functional...

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Pediatric Neurology

Background:

• Pediatric epilepsy diagnosis can be challenging.
• Resting-state functional magnetic resonance imaging (rfMRI) offers insights into brain function.
• Latency data from rfMRI may hold discriminative information.

Purpose of the Study:

• To evaluate the efficacy of CNNs using rfMRI latency data for classifying pediatric epilepsy.
• To compare the performance of CNN models against healthy controls.

Main Methods:

• Acquired preoperative rfMRI scans from 63 pediatric epilepsy patients and 259 healthy controls.
• Calculated rfMRI latency maps using voxel-wise cross-covariance.
• Trained CNN models on latency z-score maps, optimizing hyperparameters via grid-search.

Main Results:

• The best-performing CNN model achieved 85% sensitivity, 71% specificity, and an AUC of 0.86 on unseen test data.
• The model correctly classified 74% of pediatric epilepsy patients.

Conclusions:

• CNNs trained with rfMRI latency data demonstrate good performance in classifying pediatric epilepsy.
• This AI-driven approach can serve as an adjunct diagnostic tool.
• Earlier identification may improve patient referral and long-term prognosis.