TCIA Radiology Image Processing for AI and Radiomics

Area of Science:

• Medical Imaging
• Radiomics
• Artificial Intelligence

Background:

• Multi-institutional imaging repositories like The Cancer Imaging Archive (TCIA) contain heterogeneous data.
• Variations in acquisition protocols, metadata, and image quality hinder consistent radiomics and AI analyses.

Purpose of the Study:

• To develop a standardized and reproducible preprocessing framework for computed tomography (CT) imaging data.
• To enable consistent radiomics and AI analyses from heterogeneous, multi-institutional datasets.

Main Methods:

• A pipeline was developed including series filtering, DICOM-to-NIfTI conversion, orientation harmonization, voxel spacing normalization, intensity normalization, segmentation integration, and metadata validation.
• The framework is implemented in a reproducible, notebook-based format compatible with radiomics and deep learning workflows.
• TCGA-KIRC patient data from TCIA served as a representative heterogeneous dataset.

Main Results:

• The pipeline standardizes imaging data into analysis-ready volumes with consistent geometry, intensity distributions, and spatial alignment.
• Non-biological variability affecting radiomic feature stability and model performance is reduced.
• The modular design allows for task-specific adaptation of preprocessing steps.

Conclusions:

• This framework provides a foundation for robust, large-scale computational imaging studies using heterogeneous datasets.
• The developed pipeline ensures consistent radiomics and AI analyses across different imaging sources.
• The approach is generalizable to other heterogeneous imaging datasets beyond TCIA.