Application of pattern recognition techniques for classification of pediatric brain tumors by in vivo 3T ¹H‐MR spectroscopy—A multi‐center study

Abstract

Purpose3T magnetic resonance scanners have boosted clinical application of 1H‐MR spectroscopy (MRS) by offering an improved signal‐to‐noise ratio and increased spectral resolution, thereby identifying more metabolites and extending the range of metabolic information. Spectroscopic data from clinical 1.5T MR scanners has been shown to discriminate between pediatric brain tumors by applying machine learning techniques to further aid diagnosis. The purpose of this multi‐center study was to investigate the discriminative potential of metabolite profiles obtained from 3T scanners in classifying pediatric brain tumors.MethodsA total of 41 pediatric patients with brain tumors (17 medulloblastomas, 20 pilocytic astrocytomas, and 4 ependymomas) were scanned across four different hospitals. Raw spectroscopy data were processed using TARQUIN. Borderline synthetic minority oversampling technique was used to correct for the data skewness. Different classifiers were trained using linear discriminative analysis, support vector machine, and random forest techniques.ResultsSupport vector machine had the highest balanced accuracy for discriminating the three tumor types. The balanced accuracy achieved was higher than the balanced accuracy previously reported for similar multi‐center dataset from 1.5T magnets with echo time 20 to 32 ms alone.ConclusionThis study showed that 3T MRS can detect key differences in metabolite profiles for the main types of childhood tumors. Magn Reson Med 79:2359–2366, 2018. © 2017 The Authors Magnetic Resonance in Medicine published by Wiley Periodicals, Inc. on behalf of International Society for Magnetic Resonance in Medicine. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

Distinguishing between paediatric brain tumour types using multi-parametric magnetic resonance imaging and machine learning: A multi-site study

“…Two oversampling methods: data replication and SMOTE (Chawla et al, 2002;Zarinabad et al, 2018), were used to increase the ependymoma group size of training sets by 100%. The oversampled data was processed, with supervised learning, as above and results compared with no oversampling.…”

“…The ability of a learning algorithm to discriminate between classes can be quantitatively determined using methods such as 'cross-validation' (Erickson et al, 2017). Previous results have shown the ability of supervised methods to separate between tumour subtypes and high/low grade tumours using magnetic resonance spectroscopy, with 1.5T and 3T results showing 79% and 86% balanced accuracy rate (BAR), respectively (Vicente et al, 2013;Zarinabad et al, 2018).…”

“…Applications of supervised learning to oncological medical imaging have commonly utilised single measures of the tumour microenvironment (such as image texture, ADC, perfusion, or spectroscopy) to discriminate between tumour types (Fetit et al, 2018;Gill et al, 2014;Orphanidou-Vlachou et al, 2014, 2013Zarinabad et al, 2017). However, in this study, we hypothesise that combining ADC and perfusion data from tumour Region of Interest (ROI) and the whole brain, provides an increased accuracy for discriminating between low-and high-grade tumours, as well as between tumour sub-types, in comparison to ROI or whole brain measures alone.…”

Cancer metabolism in a snapshot: MRS(I)

“…Recent pattern recognition studies have been dealing with two aspects: first, diagnosis, performed mainly with SV, and second, delimitation of the abnormal area, performed with MRSI. One of the main reasons for this thematic restriction is that SV MRS can address a clinically relevant question, namely the added value of MRS to MRI for brain tumour type and grade assessment . Whatever the aspect addressed, all recent studies use a ‘targeted metabolomics’ approach, i.e.…”

“…The fact that the curated INTERPRET dataset has been made accessible to research groups allows them to develop classifiers comparing single‐centre cohorts with that dataset, for example to differentiate abscesses from tumours at 3 T . Diagnosis of paediatric brain tumours has also been the subject of two recent works at 1.5 T on a large cohort of 90 patients and at 3 T in a multicentre study on 41 patients, to distinguish among the three main paediatric brain tumours (medulloblastoma versus pilocytic astrocytoma versus ependymoma). Despite this, single‐centre studies are still reported …”