Current status of the use of machine learning and magnetic resonance imaging in the field of neuro-radiomics

Abstract

Brain tumors exhibit heterogeneous profile with anomalous hemo-dynamics. In spite of significant advances at diagnostic and therapeutic fronts in the past couple of decades, the prognosis still remains poor. Magnetic resonance imaging (MRI), which provides information about the structural and functional aspects of the tumor in a noninvasive manner, has gained a lot of popularity for evaluating brain tumors. Several studies have been proposed in the recent past that focused on quantifying the characteristics of brain tumors as seen on MRI scans in terms of various descriptors, such as shape/morphology, texture, signal strength, and temporal dynamics, and then integrating these quantitative descriptors into various diagnostic and prognostic indices. This article first presents an overview of various MRI imaging sequences, such as contrast-enhanced, dynamic susceptibility contrast, diffusion tensor imaging, and conventional MRI, used in routine clinical settings. Later, it provides a detailed overview of the current status of the use of machine learning on MRI scans, with focus on clinical applications of these imaging sequences in brain tumors, including grading, assessment of the treatment response, prediction of prognosis, and identification of molecular markers. The article also highlights current challenges and future research directions.