Abstract
Background: The incidence of Isocitrate dehydrogenase (IDH) gene mutation had closed contact with the development and prognosis of WHO grade II/III glioma. This study aims to establish and evaluate the predicting random Forest model for IDH1 gene mutation based on parenchyma and peritumoral edema ADC image texture parameters of WHO grade II/III glioma. Materials and Methods: 146 patients (77 males and 69 females) with histologically confirmed anaplastic glioma were divided into training and validation groups in a ratio of 7:3 according to the requirements of Random Forest Model. The training group consisted of 102 patients (42 IDH1 mutant and 60 wild type) and the validation group included 44 patients (18 IDH1 mutant and 26 wild type). Conventional MRI features of two independent samples (IDH1 mutant and wild type) were evaluated by the Visually Accessible Rembrandt Images (VASARI) scoring system, Texture analysis (TA) of ADC image was based on the entire tumor volume and peritumoral edema and was used as Principal component analysis (PCA) to screen texture features labels. Random forest diagnosis models (VASARI+TumorADC、VASARI+TumorADC+EdemaADC) were constructed on the basis of morphological single-factor variables, texture feature labels. Result: The diagnostic accuracy of the random forest diagnosis model (VASARI+Tumor ADC) was 71.5%, the specificity was 75.40%, and the AUC was 0.769, The model (VASARI+TumorADC +peritumoral edema ADC ) was 80.9%, 79.5% ,and 0.819 correspondingly. Conclusion: The texture parameters of peritumoral edema ADC image were non-invasive markers to predict IDH1 mutational status and they have played a certain role in improving the efficiency of diagnostic model.
Rotation invariant features based on three dimensional Gaussian Markov random fields for volumetric texture classification
