Related Experiment Video
Updated: Dec 9, 2025
10:17
Guidelines and Experience Using Imaging Biomarker Explorer IBEX for Radiomics
Published on: January 8, 2018
13.5K
Minimizing acquisition-related radiomics variability by image resampling and batch effect correction to allow for
Marta Ligero1, Olivia Jordi-Ollero2, Kinga Bernatowicz1
1Radiomics Group, Vall d'Hebron Institute of Oncology (VHIO), Hospital Universitari Vall d'Hebron, Vall d'Hebron Barcelona Hospital Campus (Spain), Barcelona, Spain.
European Radiology
|September 10, 2020
Summary
CT acquisition parameters like voxel size, slice thickness, and kernel significantly impact radiomics variability. Post-acquisition correction methods, particularly ComBat, standardize data and enhance classification accuracy for large-scale analysis.
Area of Science:
- Medical Imaging Physics
- Radiomics and Computational Pathology
- Biostatistics and Data Science
Background:
- Radiomics analysis is sensitive to variations in Computed Tomography (CT) acquisition parameters.
- Standardizing CT-image acquisition protocols is crucial for reliable, large-scale radiomics studies.
- Existing methods struggle to mitigate variability introduced by diverse CT acquisition settings.
Purpose of the Study:
- To identify specific CT acquisition parameters that contribute to radiomics variability.
- To develop and evaluate a post-acquisition CT image correction method for reducing radiomics variability.
- To improve radiomics-based classification accuracy in both phantom and clinical settings.
Main Methods:
- Prospective phantom study and multi-centric retrospective clinical study involving CT scans of liver metastases.
- Extraction of 93 first-order and texture radiomics features.
- Evaluation of variability using Intraclass Correlation Coefficients (ICCs) and exploration of voxel size resampling, ComBat, and Singular Value Decomposition (SVD) for correction.
- Comparison of robust feature counts and radiomics classification accuracy (K-means purity) before and after correction.
Main Results:
- Voxel size, reconstruction slice spacing, convolution kernel, and acquisition slice thickness were identified as significant sources of radiomics variability.
- Resampling to isometric voxels and SVD/ComBat-based corrections significantly increased the number of robust radiomics features (p < 0.05).
- ComBat correction demonstrated the highest improvement in radiomics-based classification accuracy (K-means purity from 65.98% to 73.20%) in both phantom and clinical data.
Conclusions:
- CT image post-acquisition processing and radiomics normalization using batch effect correction methods enable standardization of large-scale data.
- Correction methods effectively reduce radiomics variability stemming from slice thickness and convolution kernel.
- Standardization and correction improve the accuracy of radiomics classification for differentiating tissues and tumor types.
