Brain tumor grading diagnosis using transfer learning based on optical coherence tomography

Area of Science:

• Neurosurgery
• Artificial Intelligence
• Medical Imaging

Background:

• Accurate brain tumor identification is crucial in neurosurgery to prevent recurrence.
• Existing imaging methods have limitations in distinguishing tumor types.
• Novel techniques are needed to enhance intraoperative decision-making.

Purpose of the Study:

• To validate a transfer learning model for classifying brain tissues.
• To differentiate between normal tissue, primary central nervous system lymphoma (PCNSL), high-grade glioma (HGG), and low-grade glioma (LGG).
• To assess the clinical utility of optical coherence tomography (OCT) combined with AI for neurosurgery.

Main Methods:

• Optical coherence tomography (OCT) was used to acquire measurements from tumor specimens.
• A MobileNetV2 model, pre-trained on a large dataset, was employed for binary hierarchical classification.
• Surgeons' expertise was integrated to refine model predictions.

Main Results:

• The transfer learning model achieved robust classification accuracy for different brain tumor types.
• The AI-driven approach showed promising clinical value in distinguishing tumor tissues.
• Dynamic t-SNE visualization effectively illustrated the model's classification performance.

Conclusions:

• The validated AI model offers a novel approach to brain tumor classification in neurosurgery.
• This method has the potential to improve surgical precision and patient outcomes.
• Integrating AI with OCT provides a valuable tool for neurosurgical decision support.