Deep Learning Provides Exceptional Accuracy to ECoG-Based Functional Language Mapping for Epilepsy Surgery.

Area of Science:

• Neurosurgery
• Epilepsy Research
• Artificial Intelligence in Medicine

Background:

• Surgical resection success in epilepsy hinges on preserving critical brain regions and removing pathological tissue.
• Electrocortical stimulation mapping (ESM) is the gold standard for localizing eloquent cortex but carries risks like seizures.
• Electrocorticography-based functional mapping (ECoG-FM) is a safer alternative but has historically shown lower success rates than ESM.

Purpose of the Study:

• To develop a novel deep learning algorithm to enhance the accuracy of ECoG-FM for identifying eloquent language cortex.
• To achieve accuracy comparable to ESM using a safer ECoG-FM approach.
• To provide a stand-alone functional mapping modality for epilepsy surgery, reducing risks associated with ESM.

Main Methods:

• Development of a deep learning algorithm for signal analysis of electrocorticography data.
• Application of the algorithm to presurgical evaluation data from 11 epilepsy patients.
• Analysis of 637 electrodes using the deep learning-based approach for language region identification.

Main Results:

• The proposed deep learning algorithm achieved 83.05% accuracy in identifying language regions.
• This represents a significant 23% improvement over conventional ECoG-FM analysis, which yielded approximately 60% accuracy.
• The deep learning-powered ECoG-FM demonstrated accuracy comparable to the established ESM.

Conclusions:

• Deep learning-based ECoG-FM can effectively identify eloquent language cortex with high accuracy.
• This approach offers a safer, stand-alone alternative to ESM, mitigating risks of provoked seizures and post-surgical morbidity.
• The findings pave the way for improved functional preservation during epilepsy surgery.