Targeted metabolomics analyses for brain tumor margin assessment during surgery

Area of Science:

• Neuro-oncology
• Biomedical Spectroscopy
• Machine Learning in Medicine

Background:

• Accurate identification of residual tumor tissue during glioma surgery is critical for patient survival.
• High-Resolution Magic Angle Spinning Nuclear Magnetic Resonance (HRMAS NMR) spectroscopy aids in assessing tumor margins but faces bottlenecks in quantification time and expert dependency.
• Current machine learning methods analyzing full NMR spectra are limited by high dimensionality and noise.

Purpose of the Study:

• To investigate if focusing on informative regions within HRMAS NMR spectra can improve brain tumor margin assessment.
• To develop and validate a machine learning model for predicting metabolite quantities from targeted spectral regions.
• To enhance the automation and accuracy of intraoperative feedback for glioma surgery.

Main Methods:

• Utilized HRMAS NMR spectra normalized using the ERETIC method for calibration.
• Trained machine learning models to predict metabolite quantities from specific, annotated regions of the NMR spectra.
• Evaluated model performance for tumor margin assessment using Area Under the ROC Curve (AUC-ROC) and Area Under the Precision-Recall Curve (AUC-PR).

Main Results:

• Identifying informative spectral regions significantly improved prediction power for tumor margin assessment.
• Performance gains of up to 4.6% in AUC-ROC and 2.8% in AUC-PR were achieved.
• A novel spectral region (7.97–8.09 ppm) was identified as a potential new biomarker for glioma.

Conclusions:

• Targeted analysis of HRMAS NMR spectra offers a more powerful approach to brain tumor margin assessment compared to untargeted methods.
• This method can automate and improve the accuracy of identifying residual glioma tissue during surgery.
• The identified novel spectral region warrants further investigation as a glioma biomarker.