Quality control of 3D MRSI data in glioblastoma: Can we do without the experts?. Issue 4 (26th November 2021)

• Record Type: Journal Article
• Authors: Tensaouti, Fatima; Desmoulin, Franck; Gilhodes, Julia; Martin, Elodie; Ken, Soleakhena; Lotterie, Jean‐Albert; Noël, Georges; Truc, Gilles; Sunyach, Marie‐Pierre; Charissoux, Marie; Magné, Nicolas; Lubrano, Vincent; Péran, Patrice; Cohen‐Jonathan Moyal, Elizabeth; Laprie, Anne
• Is Part Of: Magnetic resonance in medicine. Volume 87:Issue 4(2022)
• Page Start: 1688
• Page End: 1699
• Publication Date: 2021-11-26
• Subjects: classification; glioblastoma -- quality control -- random forest -- three‐dimensional magnetic resonance spectroscopic imaging (3D MRSI); Nuclear magnetic resonance -- Periodicals; Electron paramagnetic resonance -- Periodicals; 616.07548
• DOI: 10.1002/mrm.29098
• Languages: English
• ISSNs: 0740-3194

Abstract

Abstract : Purpose: Proton magnetic resonance spectroscopic imaging (1H MRSI) is a noninvasive technique for assessing tumor metabolism. Manual inspection is still the gold standard for quality control (QC) of spectra, but it is both time‐consuming and subjective. The aim of the present study was to assess automatic QC of glioblastoma MRSI data using random forest analysis. Methods: Data for 25 patients, acquired prospectively in a preradiotherapy examination, were submitted to postprocessing with syngo.MR Spectro (VB40A; Siemens) or Java‐based magnetic resonance user interface (jMRUI) software. A total of 28 features were extracted from each spectrum for the automatic QC. Three spectroscopists also performed manual inspections, labeling each spectrum as good or poor quality. All statistical analyses, with addressing unbalanced data, were conducted with R 3.6.1 (R Foundation for Statistical Computing; https://www.r‐project.org ). Results: The random forest method classified the spectra with an area under the curve of 95.5%, sensitivity of 95.8%, and specificity of 81.7%. The most important feature for the classification was Residuum_Lipids_Versus_Fit, obtained with syngo.MR Spectro. Conclusion: The automatic QC method was able to distinguish between good‐ and poor‐quality spectra, and can be used by radiation oncologists who are not spectroscopy experts. This study revealed a novel set of MRSI signal features that are closely correlated with spectral quality.