Deep Radiogenomics of Lower-Grade Gliomas: Convolutional Neural Networks Predict Tumor Genomic Subtypes Using MR Images.
Ontology highlight
ABSTRACT: Purpose
To employ deep learning to predict genomic subtypes of lower-grade glioma (LLG) tumors based on their appearance at MRI.Materials and methods
Imaging data from The Cancer Imaging Archive and genomic data from The Cancer Genome Atlas from 110 patients from five institutions with lower-grade gliomas (World Health Organization grade II and III) were used in this study. A convolutional neural network was trained to predict tumor genomic subtype based on the MRI of the tumor. Two different deep learning approaches were tested: training from random initialization and transfer learning. Deep learning models were pretrained on glioblastoma MRI, instead of natural images, to determine if performance was improved for the detection of LGGs. The models were evaluated using area under the receiver operating characteristic curve (AUC) with cross-validation. Imaging data and annotations used in this study are publicly available.Results
The best performing model was based on transfer learning from glioblastoma MRI. It achieved AUC of 0.730 (95% confidence interval [CI]: 0.605, 0.844) for discriminating cluster-of-clusters 2 from others. For the same task, a network trained from scratch achieved an AUC of 0.680 (95% CI: 0.538, 0.811), whereas a model pretrained on natural images achieved an AUC of 0.640 (95% CI: 0.521, 0.763).Conclusion
These findings show the potential of utilizing deep learning to identify relationships between cancer imaging and cancer genomics in LGGs. However, more accurate models are needed to justify clinical use of such tools, which might be obtained using substantially larger training datasets.Supplemental material is available for this article.© RSNA, 2020.
SUBMITTER: Buda M
PROVIDER: S-EPMC8017403 | biostudies-literature |
REPOSITORIES: biostudies-literature
Similar Datasets
Project description:BACKGROUND:To investigate associations between lower-grade glioma (LGG) mRNA-based subtypes (R1-R4) and MR features. METHODS:mRNA-based subtyping was obtained from the LGG dataset in The Cancer Genome Atlas (TCGA). We identified matching patients (n?=?145) in The Cancer Imaging Archive (TCIA) who underwent MR imaging. The associations between mRNA-based subtypes and MR features were assessed. RESULTS:In the TCGA-LGG dataset, patients with the R2 subtype had the shortest median OS months (P?<?0.05). The time-dependent ROC for the R2 subtype was 0.78 for survival at 12?months, 0.76 for survival at 24?months, and 0.76 for survival at 36?months. In the TCIA-LGG dataset, 41 (23.7%) R1 subtype, 40 (23.1%) R2 subtype, 19 (11.0%) R3 subtype and 45 (26.0%) R4 subtype cases were identified. Multivariate analysis revealed that enhancing margin (ill-defined, OR: 9.985; P?=?0.003) and T1?+?C/T2 mismatch (yes, OR: 0.091; P?=?0.023) were associated with the R1 subtype (AUC: 0.708). The average accuracy of the ten-fold cross validation was 71%. Proportion of contrast-enhanced (CE) tumour (>?5%, OR: 14.733; P?<?0.001) and necrosis/cystic changes (yes, OR: 0.252; P?=?0.009) were associated with the R2 subtype (AUC: 0.832). The average accuracy of the ten-fold cross validation was 82%. Haemorrhage (yes, OR: 8.55; P?<?0.001) was positively associated with the R3 subtype (AUC: 0.689). The average accuracy of the ten-fold cross validation was 87%. Proportion of CE tumour (>?5%, OR: 0.14; P?<?0.001) was negatively associated with the R4 subtype (AUC: 0.672). The average accuracy of the ten-fold cross validation was 71%. For the prediction of the R2 subtype, the nomogram showed good discrimination and calibration. Decision curve analysis demonstrated that prediction with the R2 model was clinically useful. CONCLUSIONS:Patients with the R2 subtype had the worst prognosis. We demonstrated that MRI features can identify distinct LGG mRNA-based molecular subtypes.
| S-EPMC7322922 | biostudies-literature
Project description:ObjectiveWe aimed to identify a radiomic signature to be used as a noninvasive biomarker of prognosis in patients with lower-grade gliomas (LGGs) and to reveal underlying biological processes through comprehensive radiogenomic investigation.MethodsWe extracted 55 radiomic features from T2-weighted images of 233 patients with LGGs (training cohort: n = 85; validation cohort: n = 148). Univariate Cox regression and linear risk score formula were applied to generate a radiomic-based signature. Gene ontology analysis of highly expressed genes in the high-risk score group was conducted to establish a radiogenomic map. A nomogram was constructed for individualized survival prediction.ResultsThe six-feature radiomic signature stratified patients in the training cohort into low- or high-risk groups for overall survival (P = 0.0018). This result was successfully verified in the validation cohort (P = 0.0396). Radiogenomic analysis revealed that the prognostic radiomic signature was associated with hypoxia, angiogenesis, apoptosis, and cell proliferation. The nomogram resulted in high prognostic accuracy (C-index: 0.92, C-index: 0.70) and favorable calibration for individualized survival prediction in the training and validation cohorts.ConclusionsOur results suggest a great potential for the use of radiomic signature as a biological surrogate in providing prognostic information for patients with LGGs.
| S-EPMC6224242 | biostudies-literature
Project description:Purpose of reviewLow-grade gliomas (LGG) are a group of primary brain tumors that arise from supporting glial cells. They are characterized by a mutation in the isocitrate dehydrogenase (IDH) enzyme and include astrocytomas and oligodendrogliomas. They usually affect young adults, and their main treatment consists of surgical resection, followed by radiation and chemotherapy in selected patients. This article reviews recent research on the clinical and molecular aspects of the disease and innovative therapeutic modalities in the process.Recent findingsNewly identified clinical and molecular features are currently used in the classification of LGG and applied in treatment-planning decisions. Advanced studies on the cellular level have an advanced understanding of the metabolic effects induced by IDH mutations, offering opportunities for specific targeted therapies that may improve patient outcomes. Such findings may lead to a paradigm shift in the treatment of these tumors. Although LGG are sensitive to radiation and chemotherapy, these therapies are not curative, and patient survival remains limited, raising the need for more creative and effective interventions.
| S-EPMC7244462 | biostudies-literature
Project description:PURPOSE:Convolutional neural networks have become rapidly popular for image recognition and image analysis because of its powerful potential. In this paper, we developed a method for classifying subtypes of lung adenocarcinoma from pathological images using neural network whose that can evaluate phenotypic features from wider area to consider cellular distributions. METHODS:In order to recognize the types of tumors, we need not only to detail features of cells, but also to incorporate statistical distribution of the different types of cells. Variants of autoencoders as building blocks of pre-trained convolutional layers of neural networks are implemented. A sparse deep autoencoder which minimizes local information entropy on the encoding layer is then proposed and applied to images of size [Formula: see text]. We applied this model for feature extraction from pathological images of lung adenocarcinoma, which is comprised of three transcriptome subtypes previously defined by the Cancer Genome Atlas network. Since the tumor tissue is composed of heterogeneous cell populations, recognition of tumor transcriptome subtypes requires more information than local pattern of cells. The parameters extracted using this approach will then be used in multiple reduction stages to perform classification on larger images. RESULTS:We were able to demonstrate that these networks successfully recognize morphological features of lung adenocarcinoma. We also performed classification and reconstruction experiments to compare the outputs of the variants. The results showed that the larger input image that covers a certain area of the tissue is required to recognize transcriptome subtypes. The sparse autoencoder network with [Formula: see text] input provides a 98.9% classification accuracy. CONCLUSION:This study shows the potential of autoencoders as a feature extraction paradigm and paves the way for a whole slide image analysis tool to predict molecular subtypes of tumors from pathological features.
| S-EPMC6223755 | biostudies-other
Project description:Purpose: The present study aimed to evaluate the performance of radiomics features in the preoperative prediction of epileptic seizure following surgery in patients with LGG. Methods: This retrospective study collected 130 patients with LGG. Radiomics features were extracted from the T2-weighted MR images obtained before surgery. Multivariable Cox-regression with two nested leave-one-out cross validation (LOOCV) loops was applied to predict the prognosis, and elastic net was used in each LOOCV loop to select the predictive features. Logistic models were then built with the selected features to predict epileptic seizures at two time points. Student's t-tests were then used to compare the logistic model predicted probabilities of developing epilepsy in the epilepsy and non-epilepsy groups. The t-test was used to identify features that differentiated patients with early-onset epilepsy from their late-onset counterparts. Results: Seventeen features were selected with the two nested LOOCV loops. The index of concordance (C-index) of the Cox model was 0.683, and the logistic model predicted probabilities of seizure were significantly different between the epilepsy and non-epilepsy groups at each time point. Moreover, one feature was found to be significantly different between the patients with early- or late-onset epilepsy. Conclusion: A total of 17 radiomics features were correlated with postoperative epileptic seizures in patients with LGG and one feature was a significant predictor of the time of epilepsy onset.
| S-EPMC7360821 | biostudies-literature
Project description:BackgroundPrevious studies have shown that MR imaging features can be used to predict survival and molecular profile of glioblastoma. However, no study of a similar type has been performed on lower-grade gliomas (LGGs).MethodsPresurgical MRIs of 165 patients with diffuse low- and intermediate-grade gliomas (histological grades II and III) were scored according to the Visually Accessible Rembrandt Images (VASARI) annotations. Radiomic models using automated texture analysis and VASARI features were built to predict isocitrate dehydrogenase 1 (IDH1) mutation, 1p/19q codeletion status, histological grade, and tumor progression.ResultsInterrater analysis showed significant agreement in all imaging features scored (k = 0.703-1.000). On multivariate Cox regression analysis, no enhancement and a smooth non-enhancing margin were associated with longer progression-free survival (PFS), while a smooth non-enhancing margin was associated with longer overall survival (OS) after taking into account age, grade, tumor location, histology, extent of resection, and IDH1 1p/19q subtype. Using logistic regression and bootstrap testing evaluations, texture models were found to possess higher prediction potential for IDH1 mutation, 1p/19q codeletion status, histological grade, and progression of LGGs than VASARI features, with areas under the receiver-operating characteristic curves of 0.86 ± 0.01, 0.96 ± 0.01, 0.86 ± 0.01, and 0.80 ± 0.01, respectively.ConclusionNo enhancement and a smooth non-enhancing margin on MRI were predictive of longer PFS, while a smooth non-enhancing margin was a significant predictor of longer OS in LGGs. Textural analyses of MR imaging data predicted IDH1 mutation, 1p/19q codeletion, histological grade, and tumor progression with high accuracy.
| S-EPMC5464433 | biostudies-literature
Project description:BackgroundAs an important part of tumor immunotherapy for adjunct, therapeutic tumor vaccines have been effective against multiple solid cancers, while their efficacy against lower grade glioma (LGG) remains undefined. Immunophenotyping of tumors is an essential tool to evaluate the immune function of patients with immunodeficiency or autoimmunity. Therefore, this study aims to find the potential tumor antigen of LGG and identify the suitable population for cancer vaccination based on the immune landscape.MethodThe genomic and clinical data of 529 patients with LGG were obtained from TCGA, the mRNA_seq data of normal brain tissue were downloaded from GTEx. Differential expression gene and mutation analysis were performed to screen out potential antigens, K-M curves were carried out to investigate the correlation between the level of potential antigens and OS and DFS of patients. TIMER dataset was used to explore the correlation between genes and immune infiltrating cells. Immunophenotyping of 529 tumor samples was based on the single-sample gene sets enrichment analysis. Cibersort and Estimate algorithm were used to explore the tumor immune microenvironment characteristics in each immune subtype. Weighted gene co-expression network analysis (WGCNA) clustered immune-related genes and screened the hub genes, and pathway enrichment analyses were performed on the hub modules related to immune subtype in the WGCNA.ResultsSelecting for the mutated, up-regulated, prognosis- and immune-related genes, four potential tumor antigens were identified in LGG. They were also significantly positively associated with the antigen-presenting immune cells (APCs). Three robust immune subtypes, IS1, IS2 and IS3, represented immune status "desert", "immune inhibition", and "inflamed" respectively, which might serve as a predictive parameter. Subsequently, clinicopathological features, including the codeletion status of 1p19q, IDH mutation status, tumor mutation burden, tumor stemness, etc., were significantly different among subtypes.ConclusionFCGBP, FLNC, TLR7, and CSF2RA were potential antigens for developing cancer vaccination, and the patients in IS3 were considered the most suitable for vaccination in LGG.
| S-EPMC8369324 | biostudies-literature
Project description:BackgroundP53 mutation status is a pivotal biomarker for gliomas. Here, we developed a machine-learning model to predict p53 status in lower-grade gliomas based on radiomic features extracted from conventional magnetic resonance (MR) images.MethodsPreoperative MR images were retrospectively obtained from 272 patients with primary grade II/III gliomas. The patients were randomly allocated in a 2:1 ratio to a training (n = 180) or validation (n = 92) set. A total of 431 radiomic features were extracted from each patient. The lest absolute shrinkage and selection operator (LASSO) method was used for feature selection and radiomic signature construction. Subsequently, a machine-learning model to predict p53 status was established using the selected features and a Support Vector Machine classifier. The predictive performance of all individual features and the model was calculated using receiver operating characteristic curves in both the training and validation sets.ResultsThe p53-related radiomic signature was built using the LASSO algorithm; this procedure consisted of four first-order statistics or related wavelet features (including Maximum, Median, Minimum, and Uniformity), a shape and size-based feature (Spherical Disproportion), and ten textural features or related wavelet features (including Correlation, Run Percentage, and Sum Entropy). The prediction accuracies based on the area under the curve were 89.6% in the training set and 76.3% in the validation set, which were better than individual features.ConclusionsThese results demonstrate that MR image texture features are predictive of p53 mutation status in lower-grade gliomas. Thus, our procedure can be conveniently used to facilitate presurgical molecular pathological diagnosis.
| S-EPMC5842645 | biostudies-literature
Project description:BackgroundAlthough intrinsic immune-evasion is important in cancer proliferation, metastasis and response to treatment, it is unclear whether intrinsic immune-evasion patterns of gliomas can aid in predicting clinical prognosis and determining treatment.MethodsA total of 182 immune-evasion genes intrinsic to cancer were subjected to consensus clustering to identify immune-evasion patterns in 1421 patients with lower-grade glioma (LGG). The levels of each cancer hallmark were determined by the Gene Set Variant Analysis (GSVA) method, and immune cell infiltrations were quantified using two algorithms, the single-sample Gene Set Enrichment Analysis (ssGSEA) and the Cell-type Identification By Estimating Relative Subsets Of RNA Transcripts (CIBERSORT) methods. IEVscore was determined by a method that combined univariate Cox regression analysis, least absolute shrinkage and selection operator (LASSO) regression and principal component analysis (PCA).ResultsTranscriptional and genomic analysis showed that most immune evasion genes (IEVGs) were upregulated in LGGs, with aberrant expression driven by alterations in copy number variants (CNV). Based on the mRNA expression profiles of cancer-intrinsic IEVGs could be divided into three LGG subgroups with distinct prognosis, clinicopathological features and immune infiltrations. A combined scoring scheme designed to assess the immune-evasion levels of LGGs divided these 1421 patients into two subgroups that differed in IEVscores. LGG patients with low-IEVscore had a better prognosis, would be more likely to benefit from immune check-point inhibitors and would be more susceptible to temozolomide (TMZ) chemotherapy.ConclusionIntrinsic immune evasion in the tumor microenvironment (TME) has a crucial effect on glioma formation. Quantitatively assessing the IEV scores of individual LGG patients could enhance knowledge about the intra-glioma microenvironment and lead to the development of individualized therapeutic strategies for patients with LGG.
| S-EPMC9465887 | biostudies-literature
Project description:Purpose: Isocitrate dehydrogenase (IDH) mutations in glioma patients confer longer survival and may guide treatment decision making. We aimed to predict the IDH status of gliomas from MR imaging by applying a residual convolutional neural network to preoperative radiographic data.Experimental Design: Preoperative imaging was acquired for 201 patients from the Hospital of University of Pennsylvania (HUP), 157 patients from Brigham and Women's Hospital (BWH), and 138 patients from The Cancer Imaging Archive (TCIA) and divided into training, validation, and testing sets. We trained a residual convolutional neural network for each MR sequence (FLAIR, T2, T1 precontrast, and T1 postcontrast) and built a predictive model from the outputs. To increase the size of the training set and prevent overfitting, we augmented the training set images by introducing random rotations, translations, flips, shearing, and zooming.Results: With our neural network model, we achieved IDH prediction accuracies of 82.8% (AUC = 0.90), 83.0% (AUC = 0.93), and 85.7% (AUC = 0.94) within training, validation, and testing sets, respectively. When age at diagnosis was incorporated into the model, the training, validation, and testing accuracies increased to 87.3% (AUC = 0.93), 87.6% (AUC = 0.95), and 89.1% (AUC = 0.95), respectively.Conclusions: We developed a deep learning technique to noninvasively predict IDH genotype in grade II-IV glioma using conventional MR imaging using a multi-institutional data set. Clin Cancer Res; 24(5); 1073-81. ©2017 AACR.
| S-EPMC6051535 | biostudies-literature