Toward Trustworthy PET Radiomics: Quantitative Repeatability Analysis of ¹⁸F-FDG and ⁶⁸Ga-PSMA Features under...

Area of Science:

• Medical Imaging
• Radiomics
• Biomarker Development

Background:

• Positron Emission Tomography (PET) radiomic feature reliability is limited by inherent stochastic noise from photon counting and electronic signal processing.
• Quantifying this noise-induced variability is crucial for assessing the consistency of radiomic measurements.

Purpose of the Study:

• To evaluate measurement variability in PET radiomic features under different radiotracer distribution scenarios.
• To quantify noise-induced variability using the coefficient of variation (CV) and benchmark features based on repeatability.

Main Methods:

• Utilized an American College of Radiology-accredited phantom with uniform and non-uniform ¹⁸F-FDG and ⁶⁸Ga-PSMA tracer distributions.
• Extracted 93 radiomic features using PyRadiomics, repeating scans ten times to calculate CV.
• Classified features into Grade A (CV < 10%), Grade B (10% ≤ CV < 20%), and Grade C (CV ≥ 20%) repeatability categories.

Main Results:

• Experiment 1 (uniform distribution) yielded 22 Grade A and 14 Grade C features.
• Experiment 2 (non-uniform distribution) yielded 30 Grade A and 15 Grade C features.
• CV values showed substantial variability, ranging from 0.14% to 230% across experiments and tracers.

Conclusions:

• This study established repeatability standards for PET radiomic features by quantifying noise-induced variability.
• Findings indicate that while some features are noise-resilient, others exhibit poor repeatability.
• Development of noise-robust features is essential for trustworthy PET imaging biomarkers in clinical applications.