MRI Radiomic Features to Predict IDH1 Mutation Status in Gliomas: A Machine Learning Approach

Abstract BACKGROUND Mutation of isocitrate dehydrogenase (IDH) is not only an important landmark in the development of low-grade gliomas, but also has prognostic significance and is a potential therapeutic target. There is a high need to determinate IDH mutation status at diagnosis and during the course of therapy in a non-invasive and reliable manner. We established a machine learning approach based on a support vector machine to detect IDH mutation status in in vivo standard 1H-magnetic resonance spectroscopy (1H-MRS) at 3T with an accuracy of 88.2%, a sensitivity of 95.5% (95% CI, 77.2–99.9%), and a specificity of 75% (95% CI, 42.85–94.5%) in a prospective monocentric clinical trial. Here, the same method is applied in a retrospective cohort at 1.5T and tested for transferability. MATERIAL AND METHODS Validation cohort. The validation cohort comprised 100 patients with glioma for which standard in vivo 1H-MRS spectra had been acquired between 2002 and 2007. Standard single voxel spectroscopy had been measured at 1.5T using a PRESS sequence with a TR of 1500ms and a TE of 30ms. One sample had to be excluded due to non-malignant histology and for 15 samples the IDH mutation status was not available. Therefore, the validation cohort comprised 84 samples, of which 35 were bearing an IDH mutation in immunohistochemistry (sequencing for confirmation is outstanding). Machine learning. To transfer our method to an independent validation cohort our previously established machine learning approach was first trained on all samples of the 3T group. The trained algorithm was then applied to the data of the validation cohort. Here, among other factors the different field strengths, with which the spectra were acquired (3T vs. 1.5T) had to be considered. RESULTS 27 samples of the validation cohort had to be excluded due to poor spectra quality. Our approach correctly detected IDH mutation status in 47 of 62 patients (75.8%), although the technical conditions were significantly different from our published prospective cohort. 17 of 30 patients bearing an IDH mutation were correctly identified, while 30 of 32 wild type patients were determined successfully. CONCLUSION Our approach to detect IDH mutation status has promising application in an unselected retrospective cohort, demonstrating transferability across different technical conditions. Further investigations to improve our technique and an advanced neuropathological processing of the samples are planned.

Download Full-text

Qualitative and Quantitative MRI Analysis in IDH1 Genotype Prediction of Lower-Grade Gliomas: A Machine Learning Approach

BioMed Research International ◽

10.1155/2021/1235314 ◽

2021 ◽

Vol 2021 ◽

pp. 1-10

Author(s):

Mengqiu Cao ◽

Shiteng Suo ◽

Xiao Zhang ◽

Xiaoqing Wang ◽

Jianrong Xu ◽

...

Keyword(s):

Machine Learning ◽

Feature Extraction ◽

Feature Selection ◽

Idh1 Mutation ◽

Quantitative Mri ◽

Learning Approach ◽

Lower Grade ◽

Gray Level ◽

Qualitative And Quantitative ◽

Machine Learning Approach

Purpose. Preoperative prediction of isocitrate dehydrogenase 1 (IDH1) mutation in lower-grade gliomas (LGGs) is crucial for clinical decision-making. This study aimed to examine the predictive value of a machine learning approach using qualitative and quantitative MRI features to identify the IDH1 mutation in LGGs. Materials and Methods. A total of 102 LGG patients were allocated to training ( n = 67 ) and validation ( n = 35 ) cohorts and were subject to Visually Accessible Rembrandt Images (VASARI) feature extraction (23 features) from conventional multimodal MRI and radiomics feature extraction (56 features) from apparent diffusion coefficient maps. Feature selection was conducted using the maximum Relevance Minimum Redundancy method and 0.632+ bootstrap method. A machine learning model to predict IDH1 mutation was then established using a random forest classifier. The predictive performance was evaluated using receiver operating characteristic (ROC) curves. Results. After feature selection, the top 5 VASARI features were enhancement quality, deep white matter invasion, tumor location, proportion of necrosis, and T1/FLAIR ratio, and the top 10 radiomics features included 3 histogram features, 3 gray-level run-length matrix features, and 3 gray-level size zone matrix features and one shape feature. Using the optimal VASARI or radiomics feature sets for IDH1 prediction, the trained model achieved an area under the ROC curve (AUC) of 0.779 ± 0.001 or 0.849 ± 0.008 on the validation cohort, respectively. The fusion model that integrated outputs of both optimal VASARI and radiomics models improved the AUC to 0.879. Conclusion. The proposed machine learning approach using VASARI and radiomics features can predict IDH1 mutation in LGGs.

Download Full-text