Radiomic Applications in Oncology Imaging

Radiomic Applications in Oncology Imaging employ high-throughput quantitative analysis of routine CT, MRI and PET scans to characterise tumour phenotypes beyond visual assessment. By extracting diverse features—ranging from intensity and shape measures to sophisticated texture descriptors—radiomics generates data-driven biomarker signatures that reflect spatial heterogeneity, predict clinical outcomes and monitor treatment response. Coupled with machine learning and deep learning methods, these signatures enable risk stratification and personalised therapy across multiple cancer types. Key challenges include reproducibility of feature extraction, standardisation of imaging protocols and validation in large, multicentre cohorts. Recent advances in self-supervised modelling and robust feature assessment are addressing these challenges, paving the way for reliable integration of radiomic biomarkers into global clinical practice.

Research from Nature Portfolio

Recent studies have demonstrated the potential of large-scale self-supervised models for radiomic biomarker discovery. By training a convolutional encoder on thousands of radiographic lesions without manual labelling, researchers developed a foundation model that reduces the need for extensive annotated datasets and yields task-specific models with superior performance in biomarker extraction and classification under limited data conditions. This approach also shows enhanced stability to input variations and stronger links to underlying tumour biology. Another influential study systematically assessed the reproducibility of a comprehensive set of radiomic features across varying CT reconstruction settings. It revealed that a majority of features remain highly reproducible over different slice thicknesses, while pointing out that reconstruction algorithms should not be interchanged without careful standardisation. Such work underscores the importance of harmonised imaging protocols in ensuring reliable radiomic signatures for clinical application.

Radiomic Applications in Oncology Imaging publication trend

The graph below shows the total number of articles in radiomic applications in oncology imaging across all publications each year (not limited to Nature Index journals).

Technical terms

Radiomic feature: A quantitative descriptor from medical images capturing tumour shape, intensity or texture.

Self-supervised learning: A training method using unlabelled data to create generalisable feature representations.

Autosegmentation: Automated delineation of anatomical structures or lesions in imaging volumes using algorithms.

Area Under the Curve (AUC): A measure of a model’s discrimination ability, indicating the trade-off between sensitivity and specificity.

References

1. Foundation model for cancer imaging biomarkers. Nature Machine Intelligence (2024).
2. Reproducibility of radiomics for deciphering tumor phenotype with imaging. Scientific Reports (2016).
3. Artificial intelligence-driven radiomics study in cancer: the role of feature engineering and modeling. Military Medical Research (2023).
4. Machine learning based radiomic models outperform clinical biomarkers in predicting outcomes after immunotherapy for hepatocellular carcinoma. Journal of Hepatology (2025).