Predicting 1p/19q chromosomal deletion of brain tumors using machine learning

Cinarer G., Emiroǧlu B. G., Yurttakal A. H.

Emerging Materials Research, vol.10, no.2, pp.238-244, 2021 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 10 Issue: 2
  • Publication Date: 2021
  • Doi Number: 10.1680/jemmr.20.00350
  • Journal Name: Emerging Materials Research
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus
  • Page Numbers: pp.238-244
  • Keywords: computational studies, imaging, processing, LOW-GRADE GLIOMAS, MRI
  • Yozgat Bozok University Affiliated: Yes


© 2021 ICE Publishing: All rights reserved.Advances in molecular and genetic technologies have enabled the study of mutation and molecular changes in gliomas. The 1p/19q coding state of gliomas is important in predicting pathogenesis-based pharmacological treatments and determining innovative immunotherapeutic strategies. In this study, T1-weighted and T2-weighted fluid-attenuated inversion recovery magnetic resonance imaging (MRI) images of 121 low-grade glioma patients with biopsy-proven 1p/19q coding status and no deletion (n = 40) or co-deletion (n = 81) were used. First, regions of interests were segmented with the grow-cut algorithm. Later, 851 radiomic features including three-dimensional wavelet preprocessed and non-preprocessed ones were extracted from six different matrices such as first order, shape and texture. The extracted features were preprocessed with the synthetic minority over-sampling technique algorithm. Next, the 1p/19q decoding states of gliomas were classified using machine-learning algorithms. The best classification in the classification of glioma grades (grade II and grade III) according to 1p/19q coding status was obtained by using the logistic regression algorithm, with 93.94% accuracy and 94.74% area under the curve values. In conclusion, it was determined that non-invasive estimation of 1p/19q status from MRI images enables the selection of effective treatment strategies with early diagnosis using machine-learning algorithms without the need for surgical biopsy.