3084
Predicting molecular subgroups of medulloblastoma using a quantitative radiomics approach1MRI, the First Affiliated Hospital of Zhengzhou University, Zhengzhou, China, 2BCN Neuroimaging Center, University Medical Center Groningen, University of Groningen, Groningen, Netherlands, 3Center for Language and Brain, Shenzhen Institute of Neuroscience, Shenzhen, China, 4Neurosurgery, the First Affiliated Hospital of Zhengzhou University, Zhengzhou, China
Synopsis
Machine learning-based radiomics have been introduced in providing information on molecular biology and genomics of tumors. Here, we used features of MRI to predict molecular subgroups of medulloblastoma. MRI-based radiomics features were extracted from 37 patients with medulloblastoma (WNT = 11, SHH = 9, Group 3 = 8 , and Group 4 = 9). The molecular subgroups of medulloblastoma were classified with accepted accuracies by using support vector machine (SVM). In conclusion, MRI-based radiomics can effectively predict molecular subgroups of medulloblastoma using the machine-learning approach to benefit the treatment and prognosis of medulloblastoma.
Background and Purpose
Novel biological insights have led to consensus of classifying medulloblastoma (MB) into four distinct molecular subgroups - wingless (WNT), sonic hedgehog (SHH), Groups 3, and Group 4, which has now been incorporated in the 2016 update of the World Health Organization (WHO) classification of central nervous system (CNS) tumors. Here we showed the preliminary results of an onging project which we aim to predict the molecular subgroups of MB by using a quantitative radiomics approach based on pre-operative MRI.Methods
All patients with histological diagnosis of MB having pre-operative MRI scans including contrast-enhanced T1WI sequences and tumor tissues available for molecular profiling were considered eligible for the study. Formalin-fixed paraffin-embedded (FFPE) tissue samples were used for molecular profiling. A nanoString-based assay was conducted to categorize the molecular subgroup affiliation (WNT, SHH, Group 3, and Group 4). Thirty-seven patients with medulloblastoma (WNT = 11, SHH = 9, Group 3 = 8 , and Group 4 = 9) were enrolled, and all were underwent on a MAGNETOM Skyra 3T MR scanner (Siemens Healthcare, Erlangen, Germany) with a 32-channel head coil. A total of 1089 radiomics features were extracted from contrast-enhanced T1WI images. Then, statistically significant different features from comparision among the four groups were selected for the subsequent model training and performance evaluation by employing one way-ANOVA. After feature selection, the support vector machine (SVM) was used to train and predict molecular subgroups of MB. Finally, predictive performance was assessed by accuracy, precision, recall and F1. To access algorithm generalization, we used a 2-fold cross-validation approach.Results
The molecular subgroups of MB can be classified by important radiomics features and SVM machine-learning approach with a 70% accuracy, a 67% precision, a high recall of 92%, and a 77% F1. The molecular subgroup of SHH and Group 4 can be well classified with areas under receiver operating characteristic curves (AUC) of 0.90. AUC for classification Group 3 and Group 4, SHH and WNT, WNT and Group 4, WNT and Group 3, SHH and Group 3 were 0.77, 0.72, 0.64, 0.54, and 0.45, respectively.Discussion and Conclusion
MB is the most common malignant neoplasm of the CNS in children accounting for 20-25% of all pediatric tumors. MB consists of four molecular subgroups, including WNT, SHH, Group 3, and Group 4, with obvious differences in clinical outcomes and response to therapy. The molecular classification of MB can provide important guidance to clinical treatment, so as to improve the prognosis of patients. However, it is not routinely used in the clinical practice because of the limited technique and high cost for subgrouping. Our findings suggest a great potential to use MRI-based radiomics features and machine learning method to predict the molecular subgroups of medulloblastoma, which can benefit the treatment and prognosis of MB without increasing healthcare expenses.Acknowledgements
This study was supported by the Research Projects of Henan Higher Education (No.18A320077).References
[1] Schwalbe EC, Lindsey JC, Straughton D, et al. Rapid diagnosis of medulloblastoma molecular subgroups[J]. Clin Cancer Res, 2011, 17(7): 1883-94.
[2] Louis DN, Perry A, Reifenberger G, et al. The 2016 World Health Organization Classification of Tumors of the Central Nervous System: a summary[J]. Acta Neuropathol, 2016, 131(6): 803-20.
[3] Ramaswamy V, Remke M, Bouffet E, et al. Recurrence patterns across medulloblastoma subgroups: an integrated clinical and molecular analysis[J]. Lancet Oncol, 2013, 14(12): 1200-7.
[4] Zhukova N, Ramaswamy V, Remke M, et al. Subgroup-specific prognostic implications of TP53 mutation in medulloblastoma[J]. J Clin Oncol, 2013, 31(23): 2927-35.
[5] Northcott PA, Shih DJ, Remke M, et al. Rapid, reliable, and reproducible molecular sub-grouping of clinical medulloblastoma samples[J]. Acta Neuropathol, 2012, 123(4): 615-26.
Figures
