Radiomic features extraction (EUCAIM-SW-054_T-02-01-003)

Radiomic features extraction (EUCAIM-SW-054_T-02-01-003) extracts radiomic features from 3D medical images using the PyRadiomics library to generate quantitative imaging features for research and machine learning applications.

Key Features:

• Docker-based execution: Packaged and executed as a Docker image to provide a consistent computational environment.
• PyRadiomics integration: Leverages the PyRadiomics library to compute standardized radiomic features.
• Input formats: Processes NIfTI images natively and supports DICOM through a pre-processing conversion step.
• Parameter configuration: Allows customization of filters, bin width, resampling spacing, and normalization settings.
• Recommended bin width: A bin width of 5 is recommended for robust and reproducible feature extraction in reported experiments.
• Inhomogeneity removal: Includes inhomogeneity removal as a preprocessing option, which was reported to have minimal impact on most feature stability.
• Output format: Produces a structured CSV file containing extracted radiomic variables.

Scientific Applications:

• Reproducibility assessment: Applied to evaluate reproducibility of radiomic features from T2-weighted MR images in neuroblastoma, with normalization identified as a significant factor.
• Pediatric oncology imaging: Relevant to studies in Pediatrics, MR Imaging, Oncology, and Neuroblastic Tumors focusing on repeatability and reproducibility of imaging biomarkers.
• Quantitative imaging and machine learning: Generates feature sets suitable for downstream machine learning and quantitative imaging research.

Methodology:

Implements feature extraction with PyRadiomics on NIfTI images (DICOM via pre-processing conversion), runs in a Docker environment, allows configuration of filters, bin width, resampling spacing, normalization and optional inhomogeneity removal, and exports extracted radiomic variables to a structured CSV file.