Automated Intraoperative Visual Detection of Pediatric Epileptogenic Brain Lesions Using a Machine Learning...

Area of Science:

• Neurosurgery
• Artificial Intelligence
• Medical Imaging

Background:

• Epilepsy affects 450,000 children in the US, causing lifelong disability and risk of sudden death.
• Current surgical treatment for epilepsy is limited by visual discrimination of normal versus abnormal brain tissue.
• Inaccurate tissue resection can lead to neurological injury or failed cures.

Purpose of the Study:

• To develop and evaluate a machine learning-based segmentation model for identifying epileptogenic brain tissue.
• To improve the accuracy of surgical resection in pediatric epilepsy cases.
• To provide a benchmark for future AI model development in neurosurgery.

Main Methods:

• Collected 62 frames from live operating microscope video during pediatric epilepsy surgery.
• Trained a random forest classifier to segment images into pathological tissue or background.
• Evaluated model performance using specificity, sensitivity, and intersection over union metrics.

Main Results:

• Achieved an average specificity of 0.99, indicating high accuracy in identifying background tissue.
• Obtained a sensitivity of 0.34, suggesting room for improvement in detecting all abnormal tissue.
• Reached an intersection over union of 0.28, reflecting the overlap between predicted and actual abnormal tissue segmentation.

Conclusions:

• Machine learning classifiers show potential in avoiding misclassification of normal brain tissue during epilepsy surgery.
• The developed model provides a foundational benchmark for advancing AI in surgical decision support.
• Further research with larger datasets is warranted to enhance sensitivity and overall performance.