What is radiomics (RADIOlogics and informatics)?

What is Radiomics?

Radiomics is an emerging field that involves extracting a large number of quantitative features from medical images using advanced imaging processing and analysis tools to mine clinically actionable insights using high-dimensionality machine learning methods. 1

Definition and Core Concepts

• Radiomics transforms medical imaging from qualitative or semi-quantitative assessment to quantitative analysis by extracting thousands of imaging markers that describe the heterogeneity and spatial complexity of lesions 1, 2
• The process involves converting large volumes of clinical imaging data into numerical features that can characterize a patient's phenotype and potentially reveal tumor patterns imperceptible to the naked eye 1, 3
• Radiomics includes both "handcrafted features" (defined a priori by human operators) and data-driven features arising from deep learning neural networks 1

The Radiomics Workflow

The radiomics process follows a structured workflow:

1. Image Acquisition and Reconstruction: Obtaining high-quality imaging data from modalities such as CT, MRI, or PET 4, 5
2. Image Segmentation: Delineating regions of interest within the images 4
3. Feature Extraction: Extracting quantitative features related to shape, texture, intensity, and spatial relationships 2, 3
4. Feature Selection and Qualification: Identifying the most robust and clinically relevant features 4, 5
5. Analysis and Model Building: Developing predictive models using machine learning algorithms 2, 4

Clinical Applications

Radiomics has significant potential for:

• Prognostic Stratification: Predicting patient outcomes and disease progression 1
• Treatment Planning: Enhancing radiation therapy planning by identifying areas requiring dose escalation or reduction 1
• Response Assessment: Evaluating treatment effectiveness non-invasively 1
• Characterizing Tumor Heterogeneity: Measuring whole-tumor heterogeneity in vivo and monitoring tumor evolution over time 1

Radiomics in Specific Conditions

• Lung Cancer: Analysis of pretreatment PET/CT images using radiomics can predict local control for patients undergoing stereotactic body radiation therapy (SBRT) 1
• Esophageal Cancer: Radiomics is being actively explored for predicting treatment outcomes, longer-term prognosis, and treatment-related toxicity 1
• Cardiovascular Disease: Radiomics can identify individuals at risk of major adverse cardiovascular events (MACEs), potentially enabling more aggressive preventive measures 1

Integration with Artificial Intelligence

• Machine learning approaches leveraging radiomics help integrate large datasets and optimize the use of imaging data while minimizing interpretation bias due to interobserver variability 1
• Deep learning is increasingly being integrated with radiomics to automatically extract specific features using pretrained models, improving model performance 1, 6
• Foundation models like Radio DINO are being developed to capture rich semantic embeddings and enable robust feature extraction without manual intervention 6

Challenges and Limitations

• Variability in acquisition and reconstruction protocols necessitates standardization efforts such as the "Radiomics Quality Score" 1, 5
• The field is still evolving, with ongoing efforts to improve bias reduction techniques and conduct multicenter studies for more robust and generalizable models 4, 5
• High dimensionality of extracted features and limited sample sizes in many studies pose statistical challenges 6, 5
• Translation of radiomics research into clinical practice requires rigorous validation on independent external cohorts 3, 5

Future Directions

• Integration of radiomics with other -omics data (genomics, proteomics) for comprehensive disease characterization 3
• Development of more standardized approaches to feature extraction and analysis 5
• Increased use of artificial intelligence to automate and enhance the radiomics workflow 1, 6
• Expansion of radiomics applications beyond oncology to other medical fields 3, 4