Related Experiment Videos
PET textural features stability and pattern discrimination power for radiomics analysis: An "ad-hoc" phantoms study
L Presotto1, V Bettinardi1, E De Bernardi2
1Nuclear Medicine Unit, IRCCS San Raffaele Scientific Institute, Milano, Italy.
Summary
The fixed-bin-number (FBN) discretization method is superior for radiomic metrics (RMs) analysis in PET imaging. FBN demonstrates less dependence on SUV and processing variations, making it ideal for robust tumor characterization.
Area of Science:
- Medical Imaging
- Radiomics
- Quantitative Imaging
Background:
- Radiomics, analyzing textural features in PET images, shows promise for tumor characterization.
- Current radiomic metrics (RMs) analysis lacks standardization, necessitating investigation into processing chain impacts.
Purpose of the Study:
- To characterize the impact of discretization methods, acquisition statistics, and reconstruction algorithms on radiomic metric values.
- To investigate the influence of tumor volume and standardized uptake value (SUV) on RMs.
Main Methods:
- Calculated 39 RMs using Chang-Gung-Image-Texture-Analysis (CGITA) software with phantom data.
- Assessed the effect of categorical variables using η² and quantified discriminatory power using Cohen's d.
Main Results:
- The fixed-bin-number (FBN) discretization method outperformed fixed-bin-size (FBS) in SUV units, showing less SUV dependence (30 RMs with η² > 20% for FBS).
- FBN exhibited less influence from acquisition and reconstruction variations (37 RMs with η² < 40% for FBN vs. 20 for FBS).
- Most RMs demonstrated good discriminatory power with FBN (29/39 RMs with d > 3).
Conclusions:
- The fixed-bin-number (FBN) method is recommended for radiomic metrics analysis in PET imaging.
- A subset of 21 RMs is proposed for standardized PET radiomics analysis.