Raman spectroscopic grading of astrocytoma tissues: using soft reference information

Area of Science:

• Oncology
• Biomedical Engineering
• Spectroscopy

Background:

• Gliomas are common primary brain tumors, and precise tumor margin identification during surgery is challenging.
• Accurate intraoperative differentiation of tumor grades is crucial for effective neurosurgical treatment.

Purpose of the Study:

• To evaluate Raman spectroscopy for grading astrocytomas in moist bulk samples.
• To develop predictive classification models for intraoperative surgical guidance.

Main Methods:

• Utilized fiber-optic probe Raman spectroscopy on 80 patients' moist bulk samples.
• Developed predictive classification models, including soft logistic regression (LR), incorporating partial class memberships for borderline cases.
• Compared LR with linear discriminant analysis (LDA), noting LR's ability to utilize more spectra and patients.

Main Results:

• Soft LR models achieved approximately 85% sensitivity and specificity for normal tissue, 67% for low-grade, and 84% for high-grade astrocytoma.
• Analysis revealed unsaturated lipids differentiate normal from tumor tissue, while total lipid content indicates tumor grade.
• The high-wavenumber region (>2,800 cm⁻¹) alone was insufficient for grading.

Conclusions:

• Raman spectroscopy provides a proof of concept for grading moist astrocytoma tissues intraoperatively.
• Incorporating borderline samples enhances classifier training and testing, improving diagnostic accuracy.
• The study supports Raman spectroscopy as a valuable tool for real-time surgical guidance in neurosurgery.