Classification Tools for MRS of Cancer
Sabine Van Huffel1

1KU Leuven, Belgium

Synopsis

This lecture explains how to extract diagnostic information from the raw MR spectra possibly combined with information from other MR modalities. Starting from basic and advanced concepts in machine learning, the most important classification methods are surveyed. The application of these classifiers in assessing brain tumor heterogeneity is illustrated in a variety of case studies. In-vivo single-voxel MR Spectroscopy (MRS) as well as chemical Shift Imaging are considered, long-echo time as well as short-echo time acquisition schemes. Moreover, it is shown how to combine these measurements with other basic (T1-weighted,T2-weighted) as well as advanced (PWI, DWI, DTI, DKI) MR modalities.

Introduction

This lecture explains how to extract diagnostic information from the raw MR spectra possibly combined with information from other MR modalities. The aim is to present a variety of machine learning methods for classification of tumours with focus on automated preprocessing and diagnosis of abnormal brain tissues, in particular for the follow- up of glioblastoma multiforme (GBM). Current conventional MRI (cMRI) techniques are very useful in detecting the main features of brain tumours, such as size and location, but are insufficient in specifying the grade or evolution of the disease. Therefore, the acquisition of advanced MRI, such as perfusion weighted imaging (PWI), diffusion kurtosis imaging (DKI), and magnetic resonance spectroscopic imaging (MRSI), is necessary to provide complementary information such as blood flow, tissue organisation, and metabolism, induced by pathological changes. In the GBM experiments our aim is to discriminate and predict the evolution of patients treated with standard radiochemotherapy and immunotherapy based on conventional and advanced MRI data.

Classification Tools

Starting from basic and advanced concepts in machine learning, the most important classification methods are surveyed. Supervised (Linear Discriminant Analysis, Neural Networks and Support Vector Machines [1]), as well as unsupervised (K-means, blind source separation using nonnegativity constraints) ones are described, up to their embedding in Decision Support Systems (SpectraClassifier [2]). Their potential and limitations, in particular regarding their use in clinical practice, are investigated.

Case Studies

The application of these classifiers in assessing brain tumor heterogeneity is illustrated in a variety of case studies. In-vivo single-voxel MR Spectroscopy (MRS) as well as chemical Shift Imaging are considered, long-echo time as well as short-echo time acquisition schemes. Moreover, it is shown how to combine these measurements with other basic (T1-weighted,T2-weighted) as well as advanced (PWI, DWI, DTI, DKI) MR modalities. The latter give additional information about various tissue characteristics and can help with typing and grading of brain tumours. We first show how to extract meaningful features from the low sensitivity MR spectroscopy in-vivo data. MR spectroscopy and MR spectroscopic imaging can provide concentrations of metabolites that act as biomarkers for brain tumour characterisation. We then show how to use MR spectroscopic imaging directly in classification approaches in order to describe normal and pathologic tissue types in the brain. Starting from a patient’s (multiparametric) MR imaging and an unsupervised classifier (such as blind source separation using nonnegative matrix and tensor factorisations), allows to discover almost automatically tissue specific spectral patterns in each individual glioma patient [3]. Combined with nosologic imaging [4], tumor heterogeneity can be easily visualized, which opens a promising path towards automated brain tumour recognition. Based on metabolic characteristics, but also on the perfusion and diffusion MRI parameters, we can accurately differentiate tissue types in the brain, which can help in tumour grading and in finding regions where the tumour is more aggressive. Finally, we show how to perform supervised classification in order to predict recurrence in high grade glioma patients in their post-surgery follow-up [5][6].

Acknowledgements

Research supported by FWO project G.0869.12N, ERC Advanced Grant: BIOTENSORS (n°339804) and EU MC ITN TRANSACT 2012 (n°316679).

References

[1] J. Luts et al. (2010), ``A tutorial on support vector machine-based methods for classification problems in chemometrics'', Analytica Chimica Acta 665:129-145.

[2] S. Ortega-Martorell et al (2010), SpectraClassifier 1.0: a user friendly, automated MRS-based classifier-development system, BMC Bioinformatics 11:106.

[3] N. Sauwen et al. (2016), ``Comparison of unsupervised classification methods for brain tumor segmentation using multi-parametric MRI'', NeuroImage: Clinical, vol. 12, pp. 753-764.

[4] J. Luts J et al.(2009), ``Nosologic imaging of the brain: segmentation and classification using MRI and MRSI'', NMR in Biomedicine 22:374-390.

[5] A. Ion-Margineanu et al. (2015), ``Tumour relapse prediction using multi-parametric MR data recorded during follow-up of GBM patients'', BioMed Research International, vol. 215, Article ID 842923.

[6] A. Ion-Margineanu et al. (2017), “Classifying glioblastoma multiforme follow-up progressive vs. responsive forms using multi-parametric MRI features'', Frontiers in Neuroscience, Brain Imaging Methods, vol. 10, Jan. 2017, pp. 615.

Proc. Intl. Soc. Mag. Reson. Med. 26 (2018)