Stability of radiomic features in CT perfusion maps.

Area of Science:

• Radiomics
• Medical Imaging
• Oncology

Background:

• Radiomic parameters derived from CT perfusion (CTP) maps hold potential for predicting tumor response to chemo-radiotherapy.
• However, the robustness of these parameters is influenced by various CTP calculation factors and image discretization methods.

Purpose of the Study:

• To identify stable radiomic parameters in CTP maps that are reproducible across different calculation factors and discretization methods.
• To establish a foundation for developing reliable prognostic models for local tumor response to chemo-radiotherapy.

Main Methods:

• Analyzed pre-treatment CTP images from oropharyngeal carcinoma and non-small cell lung cancer (NSCLC) patients.
• Investigated the robustness of 315 radiomic parameters against standardizable (e.g., voxel size, image discretization) and non-standardizable (e.g., artery contouring) factors using intraclass correlation (ICC).
• Excluded parameters correlated with tumor volume and selected the highest ICC parameter from correlated stable parameters after standardization.

Main Results:

• Potentially standardizable factors, particularly voxel size (>97% instability), introduced more variability than non-standardizable factors (43-58% instability).
• Without standardization, no radiomic parameters were stable.
• After standardizing non-standardizable factors, ten radiomic parameters remained stable across both patient cohorts.

Conclusions:

• Standardization of CTP calculation factors, including voxel size, image discretization, Hounsfield unit (HU) threshold, and temporal resolution, is imperative for building reliable predictive models using CTP radiomics.
• The identified stable radiomic parameters can serve as input for future prognostic models to predict local tumor response to chemo-radiotherapy.