Improving the noninvasive classification of glioma genetic subtype with deep learning and diffusion-weighted imaging

Abstract

Background Diagnostic classification of diffuse gliomas now requires an assessment of molecular features, often including IDH-mutation and 1p19q-codeletion status. Because genetic testing requires an invasive process, an alternative noninvasive approach is attractive, particularly if resection is not recommended. The goal of this study was to evaluate the effects of training strategy and incorporation of biologically relevant images on predicting genetic subtypes with deep learning. Methods Our dataset consisted of 384 patients with newly-diagnosed gliomas who underwent preoperative MR imaging with standard anatomical and diffusion-weighted imaging, and 147 patients from an external cohort with anatomical imaging. Using tissue samples acquired during surgery, each glioma was classified into IDH-wildtype (IDHwt), IDH-mutant/1p19q-noncodeleted (IDHmut-intact), and IDH-mutant/1p19q-codeleted (IDHmut-codel) subgroups. After optimizing training parameters, top performing convolutional neural network (CNN) classifiers were trained, validated, and tested using combinations of anatomical and diffusion MRI with either a 3-class or tiered structure. Generalization to an external cohort was assessed using anatomical imaging models. Results The best model used a 3-class CNN containing diffusion-weighted imaging as an input, achieving 85.7% (95% CI:[77.1,100]) overall test accuracy and correctly classifying 95.2%, 88.9%, 60.0% of the IDHwt, IDHmut-intact, and IDHmut-codel tumors. In general, 3-class models outperformed tiered approaches by 13.5-17.5%, and models that included diffusion-weighted imaging were 5-8.8% more accurate than those that used only anatomical imaging. Conclusion Training a classifier to predict both IDH-mutation and 1p19q-codeletion status outperformed a tiered structure that first predicted IDH-mutation, then1p19q-codeletion. Including ADC, a surrogate marker of cellularity, more accurately captured differences between subgroups.

Multiparametric MRI-Based Radiomics Model for Predicting H3 K27M Mutant Status in Diffuse Midline Glioma: A Comparative Study Across Different Sequences and Machine Learning Techniques

“…Finally, the performance of deep learning algorithms was not evaluated and compared in our study. Deep learning algorithms have been widely used in glioma molecular subtype prediction (41)(42)(43)(44)(45). However, deep learning usually needs a huge amount of dataset, such as hundreds or thousands of cases, and the dataset is limited for our approach.…”

Swin Transformer Improves the IDH Mutation Status Prediction of Gliomas Free of MRI-Based Tumor Segmentation

“…It is difficult for the radiologist to distinguish glioma genotypes based on these radiographic features in clinical practice. Fortunately, leveraging the recent advances in machine learning approaches, such as deep learning (DL), SVM, decision tree, etc., IDH mutation status prediction from MRI can be operated accurately and objectively [ 11 , 12 , 13 , 14 , 15 , 16 , 17 ]. Among them, DL approaches have received the most notable attention for the reason of their outstanding performance in the molecular biomarker prediction from high-dimensional numeric information or image signal intensities [ 18 , 19 , 20 ].…”

“…Among them, DL approaches have received the most notable attention for the reason of their outstanding performance in the molecular biomarker prediction from high-dimensional numeric information or image signal intensities [ 18 , 19 , 20 ]. Besides IDH prediction [ 11 , 12 , 13 , 14 , 15 , 16 , 17 ], DL is also widely applied to 1p/19q [ 21 , 22 ], MGMT [ 23 , 24 ] prediction, etc.…”

Fully Automated MR Based Virtual Biopsy of Cerebral Gliomas

“…Over the past few years, a number of studies demonstrated the great potential of MRI-based image analyses for the tumor decoding of gliomas [ 4 ]. Here, various approaches to image analysis for predicting the genetic profile of gliomas have emerged in recent years [ 5 , 6 , 7 ]. On the one hand, radiomics features can be extracted from segmentations and be used to train machine learning algorithms [ 8 , 9 , 10 , 11 ].…”

“…Here, Guta et al were able to show that features extracted using CNN lead to better predictions for the grading of cerebral gliomas than features extracted using Radiomics [ 5 ]. Third, various networks have been established for direct image analysis of cerebral gliomas utilizing a variety of deep learning networks [ 6 , 12 ]. All approaches come with their own advantages and disadvantages.…”