M. Pilot*a (Mr), M. Tichomirowab (Dr), L. Mombaertsc (Dr), A. Huschc (Dr), M. Mittelbronnd (Prof), F. Hertele (Prof), F. Kleine Borgmannf (Dr), M. Theodoropouloug (Prof)

a Ludwig-Maximilians-Universität München, München, GERMANY ; b Service d‘endocrinoligie, Centre Hospitalier du Nord, Ettelbruck, LUXEMBOURG ; c Luxembourg Centre of Systems Biomedicine (LCSB), University of Luxembourg (UL), Esch-Sur-Alzette, LUXEMBOURG ; d Department of Cancer Research (DoCR), Luxembourg Institute of Health (LIH) and National Center of Pathology (NCP), Laboratoire national de santé (LNS), Dudelange, LUXEMBOURG ; e Service National de Neurochirurgie, Centre Hospitalier de Luxembourg, INS Group, LCSB, University of Luxembourg, Luxembourg, LUXEMBOURG ; f Luxembourg Center of Neuropathology (LCNP), Dudelange, Luxembourg and Department of Cancer Research (DoCR), Luxembourg Institute of Health (LIH), Luxembourg, LUXEMBOURG ; g Ludwig-Maximilians-Universität München, Munich, GERMANY

* pilotmichel@gmail.com

Introduction

Identifying pituitary tumour margins intraoperatively and distinguishing it from healthy tissue would greatly facilitate neurosurgical outcome. Raman spectroscopy (RS) is a non-invasive, non-destructive and quick technique that can determine the biochemical composition and discriminate between tissue types (e.g. cancerous vs. healthy) without requiring tissue processing and histopathological analysis. Herein we applied RS on functional and hormone inactive (HI) pituitary neuroendocrine tumours (PitNETs) and correlated with the histological subtypes.

Materials and Methods

Our cohort consisted of 56 snapfrozen PitNETs from patients with acromegaly (n=15: 11 somatotroph, 3 lactosomatotroph, 1 plurihormonal), Cushing’s disease (5 corticotroph), prolactinomas (2), and HI (n=34, including 1 silent corticotroph). Tissues were thawed and 527 spot measurements were taken with a fully robotized RS machine using a 785nm laser source, acquiring 1602 single values/spectrum (wavenumber scale 314-2994cm-1). Mean Raman spectra were analysed by t-SNE (t-distributed stochastic neighbour embedding) algorithm and support vector machine (SVM) classifier.

Results

The t-SNE algorithm demonstrated clear separation between functioning and HI PitNETs. Interestingly, the t-SNE clearly grouped the one 1 silent corticotroph tumour with the HI group separately from the functional corticotroph PitNETs. Dual class SVM classifier calculated between functioning and HI PitNETs a feature importance of 81% for only 10 distinct wavelengths. Dual class SVM classifier showed 90% accuracy for functioning corticotroph vs. the other PitNET groups and 98% for acromegalic vs. HI PitNETs. Multiclass SVM showed overall 85% accuracy for the classification of HI, functional corticotroph and acromegalic PitNETs. HI PitNETs misclassification was 0%.

Discussion

We show that RS can achieve very good distinction between the different tumour types and in particular between functional and HI PitNETs. These preliminary data demonstrate that intraoperative RS could be useful in identifying the different PitNETs and potentially determine its margins and distinguish it from the healthy pituitary tissue.

The author has declared no conflict of interest.