Researchers Taylor Tucker, Rachel Daniel, and Max Pearton undertook a capstone project as part of their M.S. in Data Science curriculum with the aim of ensuring correct-side treatment in Gamma Knife radiosurgery for trigeminal neuralgia.
Wrong-side brain surgery is a sentinel event and avoidable. Although such mistakes are rare, they sought to develop a machine-learning model to predict the correct treatment side using magnetic resonance imaging (MRI) scans.
The team took MRI data, explored convolutional neural networks (CNNs), and considered pre-trained models to optimize accuracy. This work seeks to improve diagnostic precision, reduce treatment errors, and streamline clinical workflows.
The study explored machine-learning approaches to prevent wrong-side radiosurgery treatment in trigeminal neuralgia patients. They evaluated four methodologies: radiometric feature models, latent vector representations, CNN classifiers, and transfer learning. While current results fall short of clinical standards, they show that machine learning can help predict the correct treatment side.
Future work should focus on integrating segmentation models to highlight signs of neurovascular compression, improving model focus and accuracy. Latent representations also offer a promising path to reduce computational demands and speed up inference. With continued development and validation, these methods could one day support neurosurgeons along with radiation oncologists and medical physicists as an automated second-opinion tool, improving patient outcomes. Safe and effective brain surgery is always the desired result.
Researchers: Taylor Tucker, Rachel Daniel, Max Pearton
Sponsors: Jason Sheehan, David Penberthy, David Schlesinger
Advisors: Abbas Kazemipour, Adam Tashman