Impact of Preprocessing Parameters in Medical Imaging-Based Radiomic Studies: A Systematic Review.

Area of Science:

• Medical Imaging
• Radiomics
• Computational Pathology

Background:

• Radiomic studies are increasingly used for diagnosis and prognosis, but face limitations.
• Radiomic features are sensitive to image preprocessing steps.
• Lack of standardized preprocessing hinders reproducibility.

Purpose of the Study:

• To review the impact of preprocessing parameters on radiomic features.
• To assess the reproducibility and reliability of radiomic features across imaging modalities.
• To identify best practices for radiomic image preprocessing.

Main Methods:

• Comprehensive literature search across four databases (PubMed, Cochrane, Embase, Scopus).
• Inclusion of studies addressing preprocessing parameter influence on feature values and model predictions.
• Exclusion of studies lacking image acquisition details; use of QUADAS-2 for bias assessment.

Main Results:

• 43 studies were selected, examining radiomics from CT, MRI, CBCT, and PET/CT.
• Key preprocessing steps influencing feature robustness include voxel resampling, normalization, and discretization.
• Voxel resampling was studied in 44% of works; discretization was commonly employed.

Conclusions:

• Harmonization of radiomic preprocessing parameters is essential for reliable and reproducible results.
• Standardized preprocessing is critical for consistent radiomic feature extraction.
• Future prospective studies are needed to validate findings.