Towards machine learning-based quantitative hyperspectral image guidance for brain tumor resection

David Black; Declan Byrne; Anna Walke; Sidong Liu; Antonio Di Ieva; Sadahiro Kaneko; Walter Stummer; Tim Salcudean; Eric Suero Molina

doi:10.1038/s43856-024-00562-3

Towards machine learning-based quantitative hyperspectral image guidance for brain tumor resection

Commun Med (Lond). 2024 Jul 4;4(1):131. doi: 10.1038/s43856-024-00562-3.

Authors

David Black¹, Declan Byrne¹, Anna Walke², Sidong Liu³, Antonio Di Ieva³, Sadahiro Kaneko⁴, Walter Stummer², Tim Salcudean¹, Eric Suero Molina^{5

6}

Affiliations

¹ Department of Electrical and Computer Engineering, University of British Columbia, Vancouver, BC, Canada.
² Department of Neurosurgery, University Hospital Münster, Münster, Germany.
³ Computational NeuroSurgery (CNS) Lab, Macquarie Medical School, Faculty of Medicine, Health and Human Sciences, Macquarie University, Sydney, NSW, Australia.
⁴ Department of Neurosurgery, Hokkaido Medical Center, National Hospital Organization, Sapporo, Japan.
⁵ Department of Neurosurgery, University Hospital Münster, Münster, Germany. eric.suero@ukmuenster.de.
⁶ Computational NeuroSurgery (CNS) Lab, Macquarie Medical School, Faculty of Medicine, Health and Human Sciences, Macquarie University, Sydney, NSW, Australia. eric.suero@ukmuenster.de.

Abstract

Background: Complete resection of malignant gliomas is hampered by the difficulty in distinguishing tumor cells at the infiltration zone. Fluorescence guidance with 5-ALA assists in reaching this goal. Using hyperspectral imaging, previous work characterized five fluorophores' emission spectra in most human brain tumors.

Methods: In this paper, the effectiveness of these five spectra was explored for different tumor and tissue classification tasks in 184 patients (891 hyperspectral measurements) harboring low- (n = 30) and high-grade gliomas (n = 115), non-glial primary brain tumors (n = 19), radiation necrosis (n = 2), miscellaneous (n = 10) and metastases (n = 8). Four machine-learning models were trained to classify tumor type, grade, glioma margins, and IDH mutation.

Results: Using random forests and multilayer perceptrons, the classifiers achieve average test accuracies of 84-87%, 96.1%, 86%, and 91% respectively. All five fluorophore abundances vary between tumor margin types and tumor grades (p < 0.01). For tissue type, at least four of the five fluorophore abundances are significantly different (p < 0.01) between all classes.

Conclusions: These results demonstrate the fluorophores' differing abundances in different tissue classes and the value of the five fluorophores as potential optical biomarkers, opening new opportunities for intraoperative classification systems in fluorescence-guided neurosurgery.

Plain language summary

Complete surgical removal of some primary brain tumors is difficult because it can be hard to distinguish the edge of the tumor. We evaluated whether the edges of tumors and the tumor type and grade can be more accurately determined if the tumor is imaged using many different wavelengths of light. We used measurements taken from the tumors of people undergoing brain tumor surgery and developed machine-learning algorithms that could predict where the edge of the tumor was. The methods could also provide information about the type and grade of the brain tumor. These classifications could potentially be used during operations to remove brain tumors more accurately and thus improve the outcome of surgery for people with brain tumors.