Impact of preprocessing methods on the Raman spectra of brain tissue

Area of Science:

• Biomedical Optics
• Neuro-oncology
• Spectroscopy

Background:

• Accurate brain tumor resection requires distinguishing cancerous from healthy tissue.
• Current surgical guidance tools have limitations, necessitating advanced techniques.
• Raman spectroscopy offers label-free identification of tumor tissues.

Purpose of the Study:

• To evaluate six different preprocessing methods for Raman spectra of brain tumor tissues.
• To determine the impact of preprocessing on spectral feature analysis and tissue classification.
• To identify optimal preprocessing strategies for intraoperative neurosurgery guidance.

Main Methods:

• Tested six preprocessing methods (polynomial fitting, morphology, derivative, commercial software, neural network) on over 900 Raman spectra from 34 fresh brain tissue samples.
• Classified samples according to CNS WHO 2021 guidelines, including gliomas, meningiomas, and breast cancer metastases.
• Utilized principal component analysis, t-SNE, and k-means clustering for data exploration and comparison.

Main Results:

• Preprocessing method significantly influenced the number, size, and spectral features of data clusters.
• Identified spectral features associated with hemoglobin, neuroglobin, carotenoids, water, protoporphyrin, proteins, and lipids.
• No single preprocessing method allowed unambiguous assignment of spectral features to specific tissue labels.

Conclusions:

• The choice of preprocessing method critically alters the appearance of Raman spectral features from brain tumors.
• Caution is advised when comparing spectral data across studies due to preprocessing variability.
• Transparency in methodology and adaptive preprocessing, such as neural networks, are crucial for in vivo neurosurgical applications.