Deep-Learning Convolutional Neural Networks Accurately Classify Genetic Mutations in Gliomas

• DOI: 10.3174/ajnr.a5667
• Publication Date: 2018-05-10T14:10:40Z
• Authors: P. Chang; J. Grinband; B.D. Weinberg; M. Bardis; M. Khy; G. Cadena; M.-Y. Su; S. Cha; C.G. Filippi; D. Bota; P. Baldi; L.M. Poisson; R. Jain; D. Chow
• Keywords: Fluid-attenuated inversion recovery; Isocitrate dehydrogenase

<h3>BACKGROUND AND PURPOSE:</h3> The World Health Organization has recently placed new emphasis on the integration of genetic information for gliomas. While tissue sampling remains criterion standard, noninvasive imaging techniques may provide complimentary insight into clinically relevant mutations. Our aim was to train a convolutional neural network independently predict underlying molecular mutation status in gliomas with high accuracy and identify most predictive features each mutation. <h3>MATERIALS METHODS:</h3> MR data were retrospectively obtained from Cancer Imaging Archives 259 patients either low- or high-grade A trained classify <i>isocitrate dehydrogenase 1</i> (<i>IDH1</i>) status, 1p/19q codeletion, <i>O6-methylguanine-DNA methyltransferase</i> (<i>MGMT</i>) promotor methylation status. Principal component analysis final layer used extract key critical successful classification. <h3>RESULTS:</h3> Classification had accuracy: <i>IDH1</i> 94%; 92%; <i>MGMT</i> 83%. Each category also associated distinctive such as definition tumor margins, T1 FLAIR suppression, extent edema, necrosis, textural features. <h3>CONCLUSIONS:</h3> results indicate that dataset, machine-learning approaches allow classification individual mutations both We show acquired an added dimensionality-reduction technique demonstrate networks are capable learning components without prior feature selection human-directed training.