PyRadiomics

PyRadiomics extracts quantitative radiomic features from medical images to standardize tumor phenotype quantification and enable interoperable, ontology-guided radiomics analyses.

Key Features:

• Standardized Methodology: Provides a consistent framework for radiomic feature extraction to enable comparability across studies and clinical settings.
• Universal Lexicon: Employs a universal lexicon to denote semantically equivalent features.
• Semantic Interoperability (FAIR): Outputs radiomic features as semantically interoperable data graph objects compliant with FAIR principles.
• Ontology-guided Workflow (O-RAW): Integrates ontology mapping and semantic meta-labels from radiation oncology and radiomics ontologies to annotate features.
• Modular Architecture: Three modules: PyRadiomics Extension processes DICOM-RT input objects into voxel intensity arrays and binary VOI masks; Feature Extraction computes radiomic features and returns a Python dictionary; Semantic Web Integration parses outputs into a W3C-compliant triple store with semantic meta-labels.
• Efficient Execution: Tested on datasets across imaging modalities and demonstrates efficient execution times on standard hardware (e.g., HP laptop with 8GB RAM).
• SPARQL Endpoint Publishing: Publishes radiomic features to a SPARQL endpoint and supports remote SPARQL querying or export to comma-separated values.

Scientific Applications:

• Tumor phenotype quantification: Provides standardized radiomic features for quantifying tumor phenotypes from medical images.
• Treatment response prediction: Supplies consistent feature sets to support development of predictive models for treatment response.
• Multicenter validation and data sharing: Enables multicenter validation studies by publishing FAIR, semantically interoperable radiomic data despite variations in imaging protocols and preprocessing.
• Clinical translation of radiomics: Supports reproducible feature extraction and interoperable data representation to aid translation into clinical workflows.

Methodology:

Implemented in Python; processes DICOM-RT objects into voxel intensity arrays and binary VOI masks; performs radiomic feature extraction and returns a Python dictionary; maps features to a universal lexicon and serializes outputs as a W3C-compliant Semantic Web triple store with semantic meta-labels from radiation oncology and radiomics ontologies; supports publication to a SPARQL endpoint and export to CSV.