Related Experiment Video
Updated: Jun 27, 2026
10:17
Guidelines and Experience Using Imaging Biomarker Explorer IBEX for Radiomics
Published on: January 8, 2018
13.2K
Mitigating Interobserver Variability in Radiomics with ComBat: A Feasibility Study
Alessia D'Anna1, Giuseppe Stella1, Anna Maria Gueli1
1Department of Physics and Astronomy "E. Majorana", University of Catania, Via Santa Sofia 64, 95123 Catania, Italy.
Journal of Imaging
|November 26, 2024
Summary
Intraobserver Features Variability (IFV) in radiomics can be reduced using ComBat harmonization. This method improves the reliability of radiomics analysis across different centers and physicians.
Area of Science:
- Radiomics
- Medical Imaging Analysis
- Biostatistics
Background:
- Radiomics studies face challenges with Intraobserver Features Variability (IFV).
- Variability arises from differences in manual or semi-automated segmentation by physicians.
- Standardizing radiomics data is crucial for multicenter studies.
Purpose of the Study:
- To investigate IFV in radiomics.
- To assess the effectiveness of ComBat harmonization in reducing IFV.
- To evaluate the impact of segmentation methods and physician experience on radiomics outcomes.
Main Methods:
- Utilized the NSCLC-Radiomics-Interobserver1 dataset with CT scans from 22 Non-Small Cell Lung Cancer patients.
- Performed manual ('vis') and semi-automated ('auto') Gross Tumor Volume (GTV) segmentations by five radiation oncologists.
- Extracted 1229 radiomic features from original and filtered images before and after ComBat harmonization.
Main Results:
- ComBat harmonization significantly reduced the percentage of statistically significant features attributed to IFV.
- For manual segmentation, significant features decreased from 83% to 34%.
- For semi-automated segmentation, significant features decreased from 75% to 33%.
Conclusions:
- ComBat harmonization effectively mitigates IFV in radiomics.
- This harmonization enhances the feasibility of multicenter radiomics research.
- Physician experience significantly impacts radiomics analysis outcomes.
