Radiomic and Machine Learning Approaches for Glioblastoma Survival Prediction

Glioblastoma remains one of the most lethal brain tumours, marked by pronounced spatial heterogeneity and limited median survival despite aggressive treatment. Radiomics harnesses high‐throughput extraction of quantitative image features from routine MRI scans to characterise tumour intensity, texture and shape beyond visual interpretation. Machine learning algorithms then integrate these radiomic descriptors with clinical and demographic data to generate predictive models for overall survival, risk stratification and potential treatment response. Recent advances have emphasised deep learning for automated feature discovery, multi‐modal imaging fusion to capture complementary biological information, and radiogenomic mapping to link image phenotypes with underlying molecular programmes. Collectively, these approaches aim to refine prognosis, guide personalised therapy and reduce reliance on invasive sampling.

Research from Nature Portfolio

Transfer‐learning frameworks have been used to derive deep radiomic signatures from multi‐modality preoperative MRI, selecting only a handful of salient deep features via regularised Cox regression to predict patient survival and stratify risk with improved accuracy over clinical factors. Multi‐channel three‐dimensional convolutional neural networks have further automated high‐level feature extraction across contrast‐enhanced, diffusion and functional MRI, combining these implicit image descriptors with key clinical variables in a support vector machine to distinguish long‐ and short‐term survivors with high classification accuracy. In parallel, fully automatic multiparametric radiomics models have systematically assessed the influence of voxel size, quantisation and intensity normalisation on feature reproducibility, yielding robust signatures for overall survival prediction and emphasising standardisation of image pre-processing for reproducible prognostic biomarkers.

Radiomic and Machine Learning Approaches for Glioblastoma Survival Prediction publication trend

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

Technical terms

Radiomics: High-throughput extraction of quantitative image features that describe tumour characteristics beyond visual inspection.

Radiogenomics: Integration of imaging features with genomic or molecular data to link non-invasive phenotypes with underlying biology.

Deep learning: A subset of machine learning employing multi-layer neural networks to automatically learn complex feature representations from data.

Convolutional neural network (CNN): A deep learning architecture particularly suited to processing grid-structured data such as medical images, using convolutional filters to capture spatial hierarchies.

Transfer learning: Technique whereby a model pretrained on one dataset or task is adapted to another, reducing the need for large labelled training cohorts.

Nomogram: A graphical tool that combines multiple prognostic variables into a single predictive model for individualized risk assessment.

References

1. MRI-derived radiomics assessing tumor-infiltrating macrophages enable prediction of immune-phenotype, immunotherapy response and survival in glioma. Biomarker Research (2024).
2. A Deep Learning-Based Radiomics Model for Prediction of Survival in Glioblastoma Multiforme. Scientific Reports (2017).
3. Multi-Channel 3D Deep Feature Learning for Survival Time Prediction of Brain Tumor Patients Using Multi-Modal Neuroimages. Scientific Reports (2019).
4. A Fully-Automatic Multiparametric Radiomics Model: Towards Reproducible and Prognostic Imaging Signature for Prediction of Overall Survival in Glioblastoma Multiforme. Scientific Reports (2017).
5. Radiogenomic Mapping of Edema/Cellular Invasion MRI-Phenotypes in Glioblastoma Multiforme. PLOS ONE (2011).
6. Artificial intelligence in neuro-oncology: advances and challenges in brain tumor diagnosis, prognosis, and precision treatment. npj Precision Oncology (2024).
7. Radiomics and radiogenomics in gliomas: a contemporary update. British Journal of Cancer (2021).