What is the role of radiomics (radiation oncology imaging) in radiotherapy for cancer patients?

Role of Radiomics in Radiotherapy

Radiomics serves as a quantitative imaging analysis tool in radiotherapy that extracts thousands of numerical features from CT, MRI, and PET scans to predict treatment outcomes, guide personalized dose planning, and identify high-risk tumor regions that require targeted intervention. 1, 2

Core Function and Mechanism

Radiomics transforms medical imaging from qualitative assessment into quantitative analysis by converting large volumes of clinical imaging data into numerical features that characterize tumor heterogeneity and spatial complexity. 3, 1 This process includes both "handcrafted features" defined by human operators and data-driven features generated through deep learning neural networks. 3, 1

The fundamental advantage is the ability to measure whole-tumor heterogeneity in vivo non-invasively and monitor tumor evolution over time, which cannot be achieved through invasive sampling methods like biopsy alone. 3, 1

Clinical Applications in Radiotherapy Planning

Treatment Outcome Prediction

• Radiomics predicts treatment response to concurrent chemoradiotherapy with AUC values ranging from 0.74 to 0.857 for pathologic complete response (pCR) in esophageal cancer. 3
• For lung cancer patients undergoing stereotactic body radiation therapy (SBRT), pretreatment 2-[18F]FDG PET/CT radiomic analysis predicts local control with sensitivity of 100% and specificity of 81% when combining two PET features. 3
• Multidimensional predictive models combining radiomics with clinical parameters demonstrate superior discriminative performance compared to clinical models alone. 3

Personalized Dose Planning and Adaptation

• Radiomics identifies high-risk regions within tumors for dose escalation that cannot be discerned based on size or intensity alone. 2
• In hypoxic tumor sub-volumes identified through radiomic analysis, dose escalation up to 86 Gy has shown feasibility with improved median overall survival by 11.2 months in [18F]FMISO-positive patients. 3
• Radiomics-based treatment response prediction enables personalized fractionation and dose adjustments tailored to individual patient needs. 2

Toxicity Prediction

• Radiomics can predict treatment-related toxicity, including rare side effects associated with concurrent chemoradiotherapy, which is essential for improving treatment outcomes and quality of life. 3
• In head and neck cancer radiotherapy, radiomic algorithms predict the risk of severe toxicities, particularly xerostomia. 4

Prognostic Stratification

• Radiomics predicts longer-term prognosis including overall survival (OS), progression-free survival (PFS), and disease-free survival (DFS), though results vary widely with C-index/AUC ranging from 0.57 to 0.822. 3
• Adding clinical parameters and genetic information to radiomic models improves predictive power for prognostic endpoints. 3

Tumor Characterization and Targeting

• Combining peritumoral and intratumoral radiomic features improves model predictive power, with radiogenomics analysis explaining associations from the immune microenvironment perspective. 3
• Radiomics enables identification of tumor subtypes non-invasively and can correlate with genetic mutations and molecular characteristics of tissues. 4

Integration with Artificial Intelligence

• Machine learning approaches leveraging radiomics integrate large datasets and minimize interpretation bias from interobserver variability. 1
• Deep learning-based radiomics using convolutional neural network (CNN) models achieve AUCs ranging from 0.635 to 0.805 for predicting neoadjuvant chemoradiotherapy response. 3

Critical Implementation Considerations

Current Limitations

• The most significant shortcoming is lack of independent external validation, which limits generalizability of radiomic models across different institutions. 3
• Variability in acquisition and reconstruction protocols across different scanners and institutions requires harmonization and standardization to minimize uncertainties within imaging data. 2
• A "Radiomics Quality Score" has been created to harmonize radiomic feature calculation methods and enable comparison between studies. 3

Practical Requirements

• Radiomics requires large volumes of clinical imaging data and high-dimensionality machine learning methods for clinically actionable insights. 3
• For co-registered planning when using MRI, sequences should be acquired in the radiotherapy planning position. 3
• Studies examining different radiotherapy and chemotherapy regimens' effects on predictive model power remain limited. 3

Future Direction

Radiomics is positioned to become a key player in personalized oncology, particularly for identifying cases where treatment de-escalation will not jeopardize therapeutic benefit while improving quality of life. 4 The integration of artificial intelligence in radiomic-driven pipelines may allow mainstreaming their use in clinical practice. 3, 1