MRI-based classification of IDH mutation and 1p/19q codeletion status of gliomas using a 2.5D hybrid multi-task convolutional neural network

Abstract

Background IDH mutation and 1p/19q codeletion status are important prognostic markers for glioma that are currently determined using invasive procedures. Our goal was to develop artificial intelligence-based methods to non-invasively determine molecular alterations from MRI. Methods Pre-operative MRI scans of 2648 glioma patients were collected from Washington University School of Medicine (WUSM; n = 835) and publicly available Brain Tumor Segmentation (BraTS; n = 378), LGG 1p/19q (n = 159), Ivy Glioblastoma Atlas Project (Ivy GAP; n = 41), The Cancer Genome Atlas (TCGA; n = 461), and the Erasmus Glioma Database (EGD; n = 774) datasets. A 2.5D hybrid convolutional neural network was proposed to simultaneously localize glioma and classify its molecular status by leveraging MRI imaging features and prior knowledge features from clinical records and tumor location. The models were trained on 223 and 348 cases for IDH and 1p/19q tasks respectively and tested on one internal (TCGA) and two external (WUSM and EGD) test sets. Results For IDH, the best-performing model achieved areas under the receiver operating characteristic (AUROC) of 0.925, 0.874, 0.933 and areas under the precision-recall curves (AUPRC) of 0.899, 0.702, 0.853 on the internal, WUSM, and EGD test sets, respectively. For 1p/19q, the best model achieved AUROCs of 0.782, 0.754, 0.842, and AUPRCs of 0.588, 0.713, 0.782, on those three data-splits, respectively. Conclusions The high accuracy of the model on unseen data showcases its generalization capabilities and suggests its potential to perform ‘virtual biopsy’ for tailoring treatment planning and overall clinical management of gliomas.

Machine Learning-Based Prediction of Glioma IDH Gene Mutation Status Using Physio-Metabolic MRI of Oxygen Metabolism and Neovascularization (A Bicenter Study)

“…The accuracy of the three studies [ 76 , 77 , 78 ] with the top performance in the independent internal testing (92.8–93.8%) was significantly higher than our best result (87.5%); the AUROC values (0.925–0.96), however, were lower than our result (0.979). Our independent internal testing cohort was quite small, at only 16 patients, which is a limitation of our study.…”

“…The three studies with the best performance in IDH classification used data from over 1000 patients in combination with advanced CNNs. Chakrabarty et al [ 76 ] used a 2.5D hybrid CNN to simultaneously locate the glioma and classify its IDH status. By leveraging MRI features and knowledge features from clinical records and tumor location, the authors achieved an accuracy and AUROC of 93.5% and 0.925 for independent internal testing and 94.1% and 0.933 for independent external testing, respectively.…”

“…The results of the independent external tests clearly demonstrate the advantages of the approach of training more complex DL models with multi-site data from many patients from public databases. Chakrabaty et al [ 76 ] and Choi et al [ 77 ] achieved accuracies of 94.1% and 87.9% and AUROC values of 0.933 and 0.94, respectively, in their independent external tests. In our study, the traditional ML models trained with cMRI data achieved the best classification results, with an accuracy of max.…”

Novel Imaging Approaches for Glioma Classification in the Era of the World Health Organization 2021 Update: A Scoping Review

“…The analyzed images were mainly institution-based (imaged locally or in a multicentric setting), with only ten studies using a public cohort for external validation [7,11,[19][20][21][22][23][24][25][26] and four studies using public datasets (such as the BraTs 2021 [27][28][29][30]) without including local data. The studies utilizing public datasets could include significantly more patients than those with local imaging data.…”

Artificial intelligence in neuroimaging of brain tumors: reality or still promise?

“…Multiple studies leveraging radiomics, deep learning, and hybrid approaches have demonstrated high performance in predicting IDH mutation status with AI [46][47][48][49][50][51][52][53][54][55], including one study which demonstrate generalizability to an external dataset [46]. Several of these studies also showed high performance in predicting 1p19q chromosomal co-deletion status [51][52][53][54].…”