Machine learning-based prediction of glioma margin from 5-ALA induced PpIX fluorescence spectroscopy

Abstract

Gliomas are infiltrative brain tumors with a margin difficult to identify. 5-ALA induced PpIX fluorescence measurements are a clinical standard, but expert-based classification models still lack sensitivity and specificity. Here a fully automatic clustering method is proposed to discriminate glioma margin. This is obtained from spectroscopic fluorescent measurements acquired with a recently introduced intraoperative set up. We describe a data-driven selection of best spectral features and show how this improves results of margin prediction from healthy tissue by comparison with the standard biomarkerbased prediction. This pilot study based on 10 patients and 50 samples shows promising results with a best performance of 77% of accuracy in healthy tissue prediction from margin tissue. Gliomas account for more than fifty percent of primitive brain tumors. They are infiltrative tumors, with a margin difficult to identify and discriminate from the surrounding healthy tissues. The world health organization (WHO) classifies gliomas in 4 grades 1 , but most studies commonly consider two separate groups: High-Grade Gliomas (HGG) and Low-Grade Gliomas (LGG). Studies have shown that in 85% cases, recurrences of HGG are localized less than 2 centimeters away from the initial tumor 2. Then, improving the extent of resection is relevant to prevent recurrence and improve life quality and expectancy 3-5. Pre-operative MRI combined with neuro-navigation is currently used in the operating theater 6,7 but shows strong limitations 8-10. 5-aminolevulinic acid (5-ALA) induced protoporphyrin IX (PpIX) fluorescence microscopy has shown its relevance in neuro-oncology 11. PpIX absorbs light at 405 nm and emits fluorescence with a main peak centered at 634 nm. This technique is the actual clinical standard for PpIX-based surgical assistance. However, its sensitivity is still limited when applied to low-density infiltrative parts of HGG 12,13 or to LGG 14. Various 5-ALA induce PpIX fluorescence spectroscopy methods have been proposed to overcome these sensitivity issues. Previous works 6,15-24 , focus on the extraction of biomarkers from the measurements, based on a priori information on the link between the biomarkers and the microenvironment of PpIX. These approaches are known as expert-based, and various biomarker models have been proposed in the literature. Quantification of PpIX concentration 15 show enhanced sensitivity either in HGG 16 or in LGG 17. Normalization procedures of biomarkers can also increase their robustness 6,18,19. Other works suggest that relevant models could be obtained based on the shape of the PpIX emission spectrum 18-26. These works show that the PpIX fluorescence emission spectral complexity in tissue is closely linked with the pathological status. However, the still unsolved origin of this complexity impairs the extraction of the best features with an expert-based related method, thus preventing the classification of measurements into relevant pathological status.

Applications of Artificial Intelligence Based on Medical Imaging in Glioma: Current State and Future Challenges

“…The SVM model attained an accuracy of 94%. Leclerc et al (144) used spectral characteristics analysis based on FS to identify healthy tissue from margin tissue in 50 samples from ten patients. A completely automated clustering technique obtained a diagnostic accuracy of 77% in predicting healthy tissues from margin tissues.…”

A Review on Advances in Intra-operative Imaging for Surgery and Therapy: Imagining the Operating Room of the Future

“…Anatomical structures, as well as surgical tools, can be automatically identified (segmented) or tracked over time to provide surgeons with decision support and context awareness. Exemplary applications include vertebrae [201] segmentation on fluoroscopy images, tissues and surgical tools tracking [202] in 3D US, vessel segmentation [203], organ segmentation and tumor margin assessment in laparoscopic imaging [204,205,206], surgical tool detection in video la- paroscopy [207], cancerous tissue [208] and organs at risk [209,210,211], panorama stitching to enlarge the field of view [212], surface reconstruction in plastic surgery [213], identification in planning radiotherapy CT, brachitherapy [214] and biopsy [81] needles segmentation in iMRI, and pyramidal tract reconstruction [215]. -Physiological parameter estimation: medical images have been used also to esteem some physical and physiological parameters not directly measurable.…”

Robust estimation of 5-ALA-induced PpIX multiple-wavelength excitation fluorescence spectroscopy for improving glioma classification

“…These fluorophores and their high variability can lead to important crosstalks with PpIX. [8][9][10] To avoid crosstalk, the overall approach is to model the baseline with everything that is not due to 5-ALA-induced PpIX. Existing approaches are effective when the emission spectral band of the baseline is far from the one of the PpIX.…”