Radiomics and Machine Learning in Glioblastoma Assessment

Radiomics involves the extraction of high-throughput quantitative features from medical images, enabling the characterisation of tumour heterogeneity beyond what is apparent to the human eye. In glioblastoma, these features—encompassing shape, texture and intensity—are combined with machine learning algorithms to predict molecular markers, stratify prognostic groups and guide personalised treatment planning. Machine learning approaches range from traditional classifiers such as support vector machines and random forests to advanced deep learning architectures that learn hierarchical representations from multiparametric MRI scans. Radiogenomics further integrates imaging phenotypes with underlying genomic data, forging a link between visual tumour characteristics and molecular subtypes. Together, these strategies aim to develop noninvasive biomarkers for key determinants of therapy response, notably O6-methylguanine-DNA methyltransferase (MGMT) promoter methylation, and to improve outcome prediction in newly diagnosed and recurrent glioblastoma. Despite encouraging results, challenges remain in standardising feature extraction, ensuring reproducibility across centres and achieving robust external validation for clinical implementation.

Research from Nature Portfolio

A novel radiogenomic classification system for MGMT promoter methylation status employs a two-stage pipeline in which deep learning-derived latent features are fused with conventional radiomic descriptors. Using a fine-tuned feature-extraction module and a rejection algorithm to isolate non-informative instances, the approach achieved classification accuracies exceeding 96 %, with sensitivity and specificity above 96 % and 97 % respectively, on a large multicentre multiparametric MRI dataset.

An optimised machine learning framework using a genetic algorithm-based wrapper and extreme gradient boosting for feature selection demonstrated high predictive performance for MGMT status. Following two-stage feature reduction, cross-validation yielded a sensitivity of 0.89, specificity of 0.97 and overall accuracy of 0.93. The model also generalised to low-grade glioma with acceptable performance, suggesting broader applicability of key radiomic features across glioma grades.

In prognostic modelling of newly diagnosed glioblastoma, an MRI-based radiomic signature comprising texture and location features successfully stratified patients into high- and low-risk groups independent of MGMT methylation. Although prediction of methylation status reached only moderate accuracy (~67 %), the radiomic risk score emerged as an independent prognostic factor and, when combined with methylation status, delineated distinct survival cohorts.

Radiomics and Machine Learning in Glioblastoma Assessment publication trend

The graph below shows the total number of articles in radiomics and machine learning in glioblastoma assessment across all publications each year (not limited to Nature Index journals).

Technical terms

Radiomics: High-throughput extraction of quantitative features from medical images.

Radiogenomics: Integration of imaging features with genomic and molecular data.

Multiparametric MRI: Combination of different MRI sequences (e.g. T1, T2, FLAIR, diffusion) to capture complementary tissue information.

MGMT promoter methylation: Epigenetic modification of the O6-methylguanine-DNA methyltransferase gene associated with chemotherapy response.

Support vector machine (SVM): Supervised classifier that separates classes by maximizing the margin between data points and a decision boundary.

Random forest: Ensemble machine learning method using multiple decision trees to improve predictive accuracy and control overfitting.

Deep learning: Machine learning approach using multilayer neural networks to learn hierarchical feature representations directly from data.

Genetic algorithm: Search heuristic inspired by natural selection used to optimise feature subsets or model parameters.

Area under the curve (AUC): Performance metric representing the ability of a classifier to distinguish between classes across all decision thresholds.

References

1. Radiogenomic classification for MGMT promoter methylation status using multi-omics fused feature space for least invasive diagnosis through mpMRI scans. Scientific Reports (2023).
2. Improving MGMT methylation status prediction of glioblastoma through optimizing radiomics features using genetic algorithm-based machine learning approach. Scientific Reports (2022).
3. Radiomics and MGMT promoter methylation for prognostication of newly diagnosed glioblastoma. Scientific Reports (2019).
4. Quality assessment of the MRI-radiomics studies for MGMT promoter methylation prediction in glioma: a systematic review and meta-analysis. European Radiology (2024).
5. Structural and advanced imaging in predicting MGMT promoter methylation of primary glioblastoma: a region of interest based analysis. BMC Cancer (2018).
6. MRI-Based Deep Learning Tools for MGMT Promoter Methylation Detection: A Thorough Evaluation. Cancers (2023).