Comparing quantitative image parameters between animal and clinical CT-scanners: a translational phantom study...

Area of Science:

• Medical Imaging
• Radiomics
• Biomedical Engineering

Background:

• Oncological research in animal models using PET/CT faces limitations in direct translation to human CT scanners due to resolution and noise.
• Radiomics, a statistical and scale-independent method, offers a potential solution to bridge this translational gap.

Purpose of the Study:

• To compare phantom-based variability of radiomics features extracted from photon counting CT (PCCT) and an animal PET/CT scanner (Albira II).
• To investigate the potential of radiomics as a translational tool from animal models to human scans.
• To assess the comparability of PCCT and animal PET/CT scanners for radiomics analysis.

Main Methods:

• Two phantoms were scanned using PCCT and the Albira II animal PET/CT scanner with varied parameters.
• Radiomics features were extracted, followed by Principal Component Analysis (PCA) and Cluster Analysis.
• Data augmentation was employed to address small dataset limitations, and a Ridge Classifier was trained and evaluated.

Main Results:

• PCA and Cluster Analysis successfully differentiated between phantom types, highlighting the comparability of both scanners.
• The Ridge Classifier achieved 93% accuracy in training but showed a generalization challenge with 62% test accuracy.
• Feature importance and cluster analysis provided insights into feature relevance and data structure.

Conclusions:

• Radiomics demonstrates significant potential as a translational tool between animal models and human diagnostics, particularly with novel photon counting techniques.
• This study represents a crucial step towards the integration of radiomics analysis into clinical practice.
• Further research is warranted to improve the generalization of radiomics models for clinical application.