Machine Learning Approaches Pertinent to Glioma Classification

Gliomas are a heterogenous group of tumors across adult and pediatric population. While intra- and inter-tumor heterogeneity pose a significant obstacle to effective therapies, it also reflects classifiable glioma traits that can be exploited for personalized and effective therapies of patients with glioma. These traits or distinct subtypes of gliomas are attributed to several quantifiable “features”, e.g., cell-of-origin, anatomical site, histopathological and radiological findings, and molecular patterns. A central tenet of machine learning – a field in computer science to learn from data by pattern recognition – is to identify these features and predict specific patterns, e.g., a subset of patients with distinct features which may share underlying molecular makeup, and thus can be leveraged for the personalized therapy. In this chapter, we will review historical aspects of glioma classification followed by the review of both classical and deep learning machine learning approaches towards glioma classification, including prevailing challenges and upcoming trends to establish machine learning aided classification in the clinical practice.