Machine learning-based approach to identify formalin-fixed paraffin-embedded glioblastoma and healthy brain tissues

Abstract

Glioblastoma is the most malignant and common high-grade brain tumor with a 14-month overall survival length. According to recent World Health Organization Central Nervous System tumor classification (2021), the diagnosis of glioblastoma requires extensive molecular genetic tests in addition to the traditional histopathological analysis of Formalin-Fixed Paraffin-Embedded (FFPE) tissues. Time-consuming and expensive molecular tests as well as the need for clinical neuropathology expertise are the challenges in the diagnosis of glioblastoma. Hence, an automated and rapid analytical detection technique for identifying brain tumors from healthy tissues is needed to aid pathologists in achieving an error-free diagnosis of glioblastoma in clinics. Here, we report on our clinical test results of Raman spectroscopy and machine learning-based glioblastoma identification methodology for a cohort of 20 glioblastoma and 18 white matter tissue samples. We used Raman spectroscopy to distinguish FFPE glioblastoma and white matter tissues applying our previously reported protocols about optimized FFPE sample preparation and Raman measurement parameters. One may analyze the composition and identify the subtype of brain tumors using Raman spectroscopy since this technique yields detailed molecule-specific information from tissues. We measured and classified the Raman spectra of neoplastic and non-neoplastic tissue sections using machine learning classifiers including support vector machine and random forest with 86.6% and 83.3% accuracies, respectively. These proof-of-concept results demonstrate that this technique might be eventually used in the clinics to assist pathologists once validated with a larger and more diverse glioblastoma cohort and improved detection accuracies.