Related Experiment Video
Updated: Oct 19, 2025
10:17
Guidelines and Experience Using Imaging Biomarker Explorer IBEX for Radiomics
Published on: January 8, 2018
13.3K
Benchmarking Various Radiomic Toolkit Features While Applying the Image Biomarker Standardization Initiative toward
Mingxi Lei1, Bino Varghese2, Darryl Hwang2
1Ming Hsieh Department of Electrical and Computer Engineering, University of Southern California, 3740 McClintock Avenue, Los Angeles, CA, 90089, USA. mingxile@usc.edu.
Journal of Digital Imaging
|September 21, 2021
Summary
Radiomic feature extraction lacks standardization across software, hindering model generalization. This study found significant variations, particularly in morphology features, indicating results may not be interchangeable.
Area of Science:
- Medical Imaging
- Radiomics
- Biomarker Discovery
Background:
- The Image Biomarkers Standardization Initiative (IBSI) aimed to standardize quantifiable imaging metrics.
- Lack of consensus on radiomic feature terminology, mathematics, and software implementation impedes generalizability.
- Inconsistent feature extraction across toolboxes prevents building or validating unified radiomic models.
Purpose of the Study:
- To assess the inter-software agreement of radiomic feature extraction using IBSI benchmarks.
- To identify variations in feature values across different radiomics software programs.
- To evaluate the consistency of feature categories (morphology, statistic/histogram, texture) across software.
Main Methods:
- Utilized IBSI-established phantom and benchmark values for comparison.
- Extracted all 173 IBSI-standardized features (11 classes) using 6 public and 1 in-house radiomics pipeline.
- Calculated relative differences to measure inter-software agreement and analyzed feature categories using Venn and UpSet diagrams.
Main Results:
- A majority of radiomic features showed good agreement across software programs.
- Morphology features exhibited relatively poor agreement and lowest quantitative assessment scores.
- Significant differences were observed in software employing different gray-level discretization methods.
Conclusions:
- Radiomic features extracted using different software programs are not consistently interchangeable.
- Morphology features are particularly susceptible to variations in extraction methods.
- Further standardization of radiomic feature extraction workflows is crucial for reliable clinical application.
