A novel fully automated MRI-based deep-learning method for classification of IDH mutation status in brain gliomas

Intrasession Reliability of Arterial Spin-Labeled MRI-Measured Noncontrast Perfusion in Glioblastoma at 3 T

Arterial spin-labeled magnetic resonance imaging can provide quantitative perfusion measurements in the brain and can be potentially used to evaluate therapy response assessment in glioblastoma (GBM). The reliability and reproducibility of this method to measure noncontrast perfusion in GBM, however, are lacking. We evaluated the intrasession reliability of brain and tumor perfusion in both healthy volunteers and patients with GBM at 3 T using pseudocontinuous labeling (pCASL) and 3D turbo spin echo (TSE) using Cartesian acquisition with spiral profile reordering (CASPR). Two healthy volunteers at a single time point and 6 newly diagnosed patients with GBM at multiple time points (before, during, and after chemoradiation) underwent scanning (total, 14 sessions). Compared with 3D GraSE, 3D TSE-CASPR generated cerebral blood flow maps with better tumor-to-normal background tissue contrast and reduced image distortions. The intraclass correlation coefficient between the 2 runs of 3D pCASL with TSE-CASPR was consistently high (≥0.90) across all normal-appearing gray matter (NAGM) regions of interest (ROIs), and was particularly high in tumors (0.98 with 95% confidence interval [CI]: 0.97-0.99). The within-subject coefficients of variation were relatively low in all normal-appearing gray matter regions of interest (3.40%-7.12%), and in tumors (4.91%). Noncontrast perfusion measured using 3D pCASL with TSE-CASPR provided robust cerebral blood flow maps in both healthy volunteers and patients with GBM with high intrasession repeatability at 3 T. This approach can be an appropriate noncontrast and noninvasive quantitative perfusion imaging method for longitudinal assessment of therapy response and management of patients with GBM.

Accuracy of Machine Learning Algorithms for the Classification of Molecular Features of Gliomas on MRI: A Systematic Literature Review and Meta-Analysis

Simple Summary

Glioma prognosis and treatment are based on histopathological characteristics and molecular profile. Following the World Health Organization (WHO) guidelines (2016), the most important molecular diagnostic markers include IDH1/2-genotype and 1p/19q codeletion status, although more recent publications also include ARTX genotype and TERT- and MGMT promoter methylation. Machine learning algorithms (MLAs), however, were described to successfully determine these molecular characteristics non-invasively by using magnetic resonance imaging (MRI) data. The aim of this review and meta-analysis was to define the diagnostic accuracy of MLAs with regard to these different molecular markers. We found high accuracies of MLAs to predict each individual molecular marker, with IDH1/2-genotype being the most investigated and the most accurate. Radiogenomics could therefore be a promising tool for discriminating genetically determined gliomas in a non-invasive fashion. Although encouraging results are presented here, large-scale, prospective trials with external validation groups are warranted.

Abstract

Treatment planning and prognosis in glioma treatment are based on the classification into low- and high-grade oligodendroglioma or astrocytoma, which is mainly based on molecular characteristics (IDH1/2- and 1p/19q codeletion status). It would be of great value if this classification could be made reliably before surgery, without biopsy. Machine learning algorithms (MLAs) could play a role in achieving this by enabling glioma characterization on magnetic resonance imaging (MRI) data without invasive tissue sampling. The aim of this study is to provide a performance evaluation and meta-analysis of various MLAs for glioma characterization. Systematic literature search and meta-analysis were performed on the aggregated data, after which subgroup analyses for several target conditions were conducted. This study is registered with PROSPERO, CRD42020191033. We identified 724 studies; 60 and 17 studies were eligible to be included in the systematic review and meta-analysis, respectively. Meta-analysis showed excellent accuracy for all subgroups, with the classification of 1p/19q codeletion status scoring significantly poorer than other subgroups (AUC: 0.748, p = 0.132). There was considerable heterogeneity among some of the included studies. Although promising results were found with regard to the ability of MLA-tools to be used for the non-invasive classification of gliomas, large-scale, prospective trials with external validation are warranted in the future.

MRI-Based Deep-Learning Method for Determining Glioma MGMT Promoter Methylation Status

BACKGROUND AND PURPOSE

O6-Methylguanine-DNA methyltransferase (MGMT) promoter methylation confers an improved prognosis and treatment response in gliomas. We developed a deep learning network for determining MGMT promoter methylation status using T2 weighted Images (T2WI) only.

MATERIALS AND METHODS

Brain MR imaging and corresponding genomic information were obtained for 247 subjects from The Cancer Imaging Archive and The Cancer Genome Atlas. One hundred sixty-three subjects had a methylated MGMT promoter. A T2WI-only network (MGMT-net) was developed to determine MGMT promoter methylation status and simultaneous single-label tumor segmentation. The network was trained using 3D-dense-UNets. Three-fold cross-validation was performed to generalize the performance of the networks. Dice scores were computed to determine tumor-segmentation accuracy.

RESULTS

The MGMT-net demonstrated a mean cross-validation accuracy of 94.73% across the 3 folds (95.12%, 93.98%, and 95.12%, [SD, 0.66%]) in predicting MGMT methylation status with a sensitivity and specificity of 96.31% [SD, 0.04%] and 91.66% [SD, 2.06%], respectively, and a mean area under the curve of 0.93 [SD, 0.01]. The whole tumor-segmentation mean Dice score was 0.82 [SD, 0.008].

CONCLUSIONS

We demonstrate high classification accuracy in predicting MGMT promoter methylation status using only T2WI. Our network surpasses the sensitivity, specificity, and accuracy of histologic and molecular methods. This result represents an important milestone toward using MR imaging to predict prognosis and treatment response.

Fully automated hybrid approach to predict the IDH mutation status of gliomas via deep learning and radiomics

BACKGROUND

Glioma prognosis depends on isocitrate dehydrogenase (IDH) mutation status. We aimed to predict the IDH status of gliomas from preoperative MR images using a fully automated hybrid approach with convolutional neural networks (CNNs) and radiomics.

METHODS

We reviewed 1166 preoperative MR images of gliomas (grades II-IV) from Severance Hospital (n = 856), Seoul National University Hospital (SNUH; n = 107), and The Cancer Imaging Archive (TCIA; n = 203). The Severance set was subdivided into the development (n = 727) and internal test (n = 129) sets. Based on T1 postcontrast, T2, and fluid-attenuated inversion recovery images, a fully automated model was developed that comprised a CNN for tumor segmentation (Model 1) and CNN-based classifier for IDH status prediction (Model 2) that uses a hybrid approach based on 2D tumor images and radiomic features from 3D tumor shape and loci guided by Model 1. The trained model was tested on internal (a subset of the Severance set) and external (SNUH and TCIA) test sets.

RESULTS

The CNN for tumor segmentation (Model 1) achieved a dice coefficient of 0.86-0.92 across datasets. Our hybrid model achieved accuracies of 93.8%, 87.9%, and 78.8%, with areas under the receiver operating characteristic curves of 0.96, 0.94, and 0.86 and areas under the precision-recall curves of 0.88, 0.82, and 0.81 in the internal test, SNUH, and TCIA sets, respectively.

CONCLUSIONS

Our fully automated hybrid model demonstrated the potential to be a highly reproducible and generalizable tool across different datasets for the noninvasive prediction of the IDH status of gliomas.

Differentiation of Glioma Mimicking Encephalitis and Encephalitis Using Multiparametric MR-Based Deep Learning

Background

Computational aid for diagnosis based on convolutional neural network (CNN) is promising to improve clinical diagnostic performance. Therefore, we applied pretrained CNN models in multiparametric magnetic resonance (MR) images to classify glioma mimicking encephalitis and encephalitis.

Methods

A data set containing 3064 MRI brain images from 164 patients with a final diagnosis of glioma (n = 56) and encephalitis (n = 108) patients and divided into training and testing sets. We applied three MRI modalities [fluid attenuated inversion recovery (FLAIR), contrast enhanced-T1 weighted imaging (CE-T1WI) and T2 weighted imaging (T2WI)] as the input data to build three pretrained deep CNN models (Alexnet, ResNet-50, and Inception-v3), and then compared their classification performance with radiologists' diagnostic performance. These models were evaluated by using the area under the receiver operator characteristic curve (AUC) of a five-fold cross-validation and the accuracy, sensitivity, specificity were analyzed.

Results

The three pretrained CNN models all had AUC values over 0.9 with excellent performance. The highest classification accuracy of 97.57% was achieved by the Inception-v3 model based on the T2WI data. In addition, Inception-v3 performed statistically significantly better than the Alexnet architecture (p<0.05). For Inception-v3 and ResNet-50 models, T2WI offered the highest accuracy, followed by CE-T1WI and FLAIR. The performance of Inception-v3 and ResNet-50 had a significant difference with radiologists (p<0.05), but there was no significant difference between the results of the Alexnet and those of a more experienced radiologist (p >0.05).

Conclusions

The pretrained CNN models can automatically and accurately classify these two diseases and further help to improving clinical diagnostic performance.

Designing deep learning studies in cancer diagnostics

The number of publications on deep learning for cancer diagnostics is rapidly increasing, and systems are frequently claimed to perform comparable with or better than clinicians. However, few systems have yet demonstrated real-world medical utility. In this Perspective, we discuss reasons for the moderate progress and describe remedies designed to facilitate transition to the clinic. Recent, presumably influential, deep learning studies in cancer diagnostics, of which the vast majority used images as input to the system, are evaluated to reveal the status of the field. By manipulating real data, we then exemplify that much and varied training data facilitate the generalizability of neural networks and thus the ability to use them clinically. To reduce the risk of biased performance estimation of deep learning systems, we advocate evaluation in external cohorts and strongly advise that the planned analyses, including a predefined primary analysis, are described in a protocol preferentially stored in an online repository. Recommended protocol items should be established for the field, and we present our suggestions.

A Quantitative Assessment of Pre-Operative MRI Reports in Glioma Patients: Report Metrics and IDH Prediction Ability

Objectives

To measure the metrics of glioma pre-operative MRI reports and build IDH prediction models.

Methods

Pre-operative MRI reports of 144 glioma patients in a single institution were collected retrospectively. Words were transformed to lowercase letters. White spaces, punctuations, and stop words were removed. Stemming was performed. A word cloud method applied to processed text matrix visualized language behavior. Spearman's rank correlation assessed the correlation between the subjective descriptions of the enhancement pattern. The T1-contrast images associated with enhancement descriptions were selected. The keywords associated with IDH status were evaluated by χ2 value ranking. Random forest, k-nearest neighbors and Support Vector Machine algorithms were used to train models based on report features and age. All statistical analysis used two-tailed test with significance at p <.05.

Results

Longer word counts occurred in reports of older patients, higher grade gliomas, and wild type IDH gliomas. We identified 30 glioma enhancement descriptions, eight of which were commonly used: peripheral, heterogeneous, irregular, nodular, thick, rim, large, and ring. Five of eight patterns were correlated. IDH mutant tumors were characterized by words related to normal, symmetric or negative findings. IDH wild type tumors were characterized words by related to pathological MR findings like enhancement, necrosis and FLAIR foci. An integrated KNN model based on report features and age demonstrated high-performance (AUC: 0.89, 95% CI: 0.88-0.90).

Conclusion

Report length depended on age, glioma grade, and IDH status. Description of glioma enhancement was varied. Report descriptions differed for IDH wild and mutant gliomas. Report features can be used to predict glioma IDH status.

A novel fully automated MRI-based deep-learning method for classification of 1p/19q co-deletion status in brain gliomas

BACKGROUND

One of the most important recent discoveries in brain glioma biology has been the identification of the isocitrate dehydrogenase (IDH) mutation and 1p/19q co-deletion status as markers for therapy and prognosis. 1p/19q co-deletion is the defining genomic marker for oligodendrogliomas and confers a better prognosis and treatment response than gliomas without it. Our group has previously developed a highly accurate deep-learning network for determining IDH mutation status using T2-weighted (T2w) MRI only. The purpose of this study was to develop a similar 1p/19q deep-learning classification network.

METHODS

Multiparametric brain MRI and corresponding genomic information were obtained for 368 subjects from The Cancer Imaging Archive and The Cancer Genome Atlas. 1p/19 co-deletions were present in 130 subjects. Two-hundred and thirty-eight subjects were non-co-deleted. A T2w image-only network (1p/19q-net) was developed to perform 1p/19q co-deletion status classification and simultaneous single-label tumor segmentation using 3D-Dense-UNets. Three-fold cross-validation was performed to generalize the network performance. Receiver operating characteristic analysis was also performed. Dice scores were computed to determine tumor segmentation accuracy.

RESULTS

1p/19q-net demonstrated a mean cross-validation accuracy of 93.46% across the 3 folds (93.4%, 94.35%, and 92.62%, SD = 0.8) in predicting 1p/19q co-deletion status with a sensitivity and specificity of 0.90 ± 0.003 and 0.95 ± 0.01, respectively and a mean area under the curve of 0.95 ± 0.01. The whole tumor segmentation mean Dice score was 0.80 ± 0.007.

CONCLUSION

We demonstrate high 1p/19q co-deletion classification accuracy using only T2w MR images. This represents an important milestone toward using MRI to predict glioma histology, prognosis, and response to treatment.

The Value of Enhanced MR Radiomics in Estimating the IDH1 Genotype in High-Grade Gliomas

Background

The prognosis of IDH1-mutant glioma is significantly better than that of wild-type glioma, and the preoperative identification of IDH mutations in glioma is essential for the formulation of surgical procedures and prognostic assessment.

Purpose

To explore the value of a radiomic model based on preoperative-enhanced MR images in the assessment of the IDH1 genotype in high-grade glioma.

Materials and Methods

A retrospective analysis was performed on 182 patients with high-grade glioma confirmed by surgical pathology between December 2012 and January 2019 in our hospital with complete preoperative brain-enhanced MR images, including 79 patients with an IDH1 mutation (45 patients with WHO grade III and 34 patients with WHO grade IV) and 103 patients with wild-type IDH1 (33 patients with WHO grade III and 70 patients with WHO grade IV). Patients were divided into a primary dataset and a validation dataset at a ratio of 7 : 3 using a stratified random sampling; radiomic features were extracted using A.K. (Analysis Kit, GE Healthcare) software and were initially reduced using the Kruskal-Wallis and Spearman analyses. Lasso was finally conducted to obtain the optimized subset of the feature to build the radiomic model, and the model was then tested with cross-validation. ROC (receiver operating characteristic curve) analysis was performed to evaluate the performance of the model.

Results

The radiomic model showed good discrimination in both the primary dataset (AUC = 0.87, 95% CI: 0.754 to 0.855, ACC = 0.798, sensitivity = 85.5%, specificity = 75.4%, positive predictive value = 0.734, and negative predictive value = 0.867) and the validation dataset (AUC = 0.86, 95% CI: 0.690 to 0.913, ACC = 0.789, sensitivity = 91.3%, specificity = 69.0%, positive predictive value = 0.700, and negative predictive value = 0.909).

Conclusion

The radiomic model, based on the preoperative-enhanced MR, can effectively predict the IDH1 genotype in high-grade glioma.

Machine Learning-Based Radiomics Predicting Tumor Grades and Expression of Multiple Pathologic Biomarkers in Gliomas

BACKGROUND

The grading and pathologic biomarkers of glioma has important guiding significance for the individual treatment. In clinical, it is often necessary to obtain tumor samples through invasive operation for pathological diagnosis. The present study aimed to use conventional machine learning algorithms to predict the tumor grades and pathologic biomarkers on magnetic resonance imaging (MRI) data.

METHODS

The present study retrospectively collected a dataset of 367 glioma patients, who had pathological reports and underwent MRI scans between October 2013 and March 2019. The radiomic features were extracted from enhanced MRI images, and three frequently-used machine-learning models of LC, Support Vector Machine (SVM), and Random Forests (RF) were built for four predictive tasks: (1) glioma grades, (2) Ki67 expression level, (3) GFAP expression level, and (4) S100 expression level in gliomas. Each sub dataset was split into training and testing sets at a ratio of 4:1. The training sets were used for training and tuning models. The testing sets were used for evaluating models. According to the area under curve (AUC) and accuracy, the best classifier was chosen for each task.

RESULTS

The RF algorithm was found to be stable and consistently performed better than Logistic Regression and SVM for all the tasks. The RF classifier on glioma grades achieved a predictive performance (AUC: 0.79, accuracy: 0.81). The RF classifier also achieved a predictive performance on the Ki67 expression (AUC: 0.85, accuracy: 0.80). The AUC and accuracy score for the GFAP classifier were 0.72 and 0.81. The AUC and accuracy score for S100 expression levels are 0.60 and 0.91.

CONCLUSION

The machine-learning based radiomics approach can provide a non-invasive method for the prediction of glioma grades and expression levels of multiple pathologic biomarkers, preoperatively, with favorable predictive accuracy and stability.

XGBoost Improves Classification of MGMT Promoter Methylation Status in IDH1 Wildtype Glioblastoma

Approximately 96% of patients with glioblastomas (GBM) have IDH1 wildtype GBMs, characterized by extremely poor prognosis, partly due to resistance to standard temozolomide treatment. O6-Methylguanine-DNA methyltransferase (MGMT) promoter methylation status is a crucial prognostic biomarker for alkylating chemotherapy resistance in patients with GBM. However, MGMT methylation status identification methods, where the tumor tissue is often undersampled, are time consuming and expensive. Currently, presurgical noninvasive imaging methods are used to identify biomarkers to predict MGMT methylation status. We evaluated a novel radiomics-based eXtreme Gradient Boosting (XGBoost) model to identify MGMT promoter methylation status in patients with IDH1 wildtype GBM. This retrospective study enrolled 53 patients with pathologically proven GBM and tested MGMT methylation and IDH1 status. Radiomics features were extracted from multimodality MRI and tested by F-score analysis to identify important features to improve our model. We identified nine radiomics features that reached an area under the curve of 0.896, which outperformed other classifiers reported previously. These features could be important biomarkers for identifying MGMT methylation status in IDH1 wildtype GBM. The combination of radiomics feature extraction and F-core feature selection significantly improved the performance of the XGBoost model, which may have implications for patient stratification and therapeutic strategy in GBM.

A Nomogram Modeling 11C-MET PET/CT and Clinical Features in Glioma Helps Predict IDH Mutation

Purpose: We developed a ¹¹C-Methionine positron emission tomography/computed tomography (¹¹C-MET PET/CT)-based nomogram model that uses easy-accessible imaging and clinical features to achieve reliable non-invasive isocitrate dehydrogenase (IDH)-mutant prediction with strong clinical translational capability.

Methods: One hundred and ten patients with pathologically proven glioma who underwent pretreatment ¹¹C-MET PET/CT were retrospectively reviewed. IDH genotype was determined by IDH1 R132H immunohistochemistry staining. Maximum, mean and peak tumor-to-normal brain tissue (TNRmax, TNRmean, TNRpeak), metabolic tumor volume (MTV), total lesion methionine uptake (TLMU), and standard deviation of SUV (SUV_SD) of the lesions on MET PET images were obtained via a dedicated workstation (Siemens. syngo.via). Univariate and multivariate logistic regression models were used to identify the predictive factors for IDH mutation. Nomogram and calibration plots were further performed.

Results: In the entire population, TNRmean, TNRmax, TNRpeak, and SUV_SD of IDH-mutant glioma patients were significantly lower than these values of IDH wildtype. Receiver operating characteristic (ROC) analysis suggested SUV_SD had the best performance for IDH-mutant discrimination (AUC = 0.731, cut-off ≤ 0.29, p < 0.001). All pairs of the ¹¹C-MET PET metrics showed linear associations by Pearson correlation coefficients between 0.228 and 0.986. Multivariate analyses demonstrated that SUV_SD (>0.29 vs. ≤ 0.29 OR: 0.053, p = 0.010), dichotomized brain midline structure involvement (no vs. yes OR: 26.52, p = 0.000) and age (≤ 45 vs. >45 years OR: 3.23, p = 0.023), were associated with a higher incidence of IDH mutation. The nomogram modeling showed good discrimination, with a C-statistics of 0.866 (95% CI: 0.796–0.937) and was well-calibrated.

Conclusions:¹¹C-Methionine PET/CT imaging features (SUV_SD and the involvement of brain midline structure) can be conveniently used to facilitate the pre-operative prediction of IDH genotype. The nomogram model based on ¹¹C-Methionine PET/CT and clinical age features might be clinically useful in non-invasive IDH mutation status prediction for untreated glioma patients.

The MRI Features and Prognosis of Gliomas Associated With IDH1 Mutation: A Single Center Study in Southwest China

Purpose: To investigate the associations of MRI radiological features and prognosis of glioma with the status of isocitrate dehydrogenase 1 (IDH1).

Material and Methods: A total of 116 patients with gliomas were retrospectively recruited from January 2013 to December 2015. All patients were undergone routine MRI (T1WI, T2WI, T2-FLAIR) scanning and contrast-enhanced MRI T1WI before surgery. The following imaging features were included: tumor location, diameter, the pattern of growth, boundary, the degree of enhancement, mass effect, edema, cross the middle line, under the ependyma. χ² and Fisher's exact probability tests were used to determine the significance of associations between MRI features and IDH1 mutation of glioma. The survival distributions were estimated using Kaplan-Meier compared by Log-rank test. Univariate and multivariate analyses were performed using Cox regression.

Results: Gliomas with IDH1 mutant were significantly more likely to exhibit homogeneous signal intensity (p = 0.009) on non-contrast MRI protocols and less contrast enhancement (p = 0.000) on contrast enhanced T1WI. IDH1 mutant type glioma was more inclined to cross the midline to invade contralateral hemisphere (p = 0.001). The overall survival between IDH1 mutated and wild type glioma were significantly different (p = 0.000), age ≤ 40 (p = 0.003), KPS scores > 80 before operation (p = 0.000) and low grade glioma (p = 0.000).

Conclusions: Our results suggest IDH1 mutant in gliomas is more likely to exhibit homogeneous signal intensity, less contrast enhancement and more inclined to cross the midline. Patients with IDH1 mutated, age ≤ 40, KPS scores > 80 before operation and low-grade glioma may have a longer life and better prognosis.