Intraoperative brain tumor classification via laser-induced fluorescence spectroscopy and machine learning.

Area of Science:

• Neurosurgery
• Biomedical Engineering
• Spectroscopy

Background:

• Accurate intraoperative identification of brain tumor tissues and their borders is crucial for optimizing neurosurgical resection.
• Current methods may involve delays or tissue manipulation, impacting surgical workflow and patient outcomes.

Purpose of the Study:

• To evaluate the efficacy of a novel, nondestructive laser-based endogenous fluorescence spectroscopy device, TumorID, for rapid intraoperative classification of brain tumor specimens.
• To assess the performance of a machine learning algorithm in differentiating glioma, meningioma, pituitary adenoma, and nonneoplastic brain tissue using TumorID data.

Main Methods:

• TumorID utilizes a 100-mW, 405-nm laser for rapid (0.5 seconds), dye-free, nondestructive ex vivo scanning of tissue specimens.
• A support vector machine algorithm was trained on data from 46 patients to classify tissue types.
• Statistical analysis involved generalized estimating equations focusing on fluorophore emission regions (NADH, FAD, porphyrins).

Main Results:

• The machine learning model achieved a high multiclass area under the receiver operating characteristic curve of 0.809 ± 0.002.
• Neutral porphyrins significantly contributed to the model's classification power (p < 0.001).
• The system demonstrated rapid and accurate classification of various pathologies and surrounding brain tissue.

Conclusions:

• The TumorID device shows promise for real-time intraoperative tissue diagnostics in neurosurgery.
• Its ability to rapidly differentiate pathologies can aid surgical decision-making and border delineation.
• This technology has the potential to enhance the completeness of tumor resection and improve patient outcomes.