Comparative Analysis of Repeatability in CT Radiomics and Dosiomics Features under Image Perturbation: A Study in...

Area of Science:

• Medical Imaging
• Radiotherapy Oncology
• Quantitative Imaging

Background:

• Cervical cancer treatment relies on accurate imaging for planning and response assessment.
• Radiomics and dosiomics offer quantitative insights but their repeatability needs evaluation.
• Image perturbation techniques can assess feature robustness.

Purpose of the Study:

• To evaluate the repeatability of radiomics and dosiomics features in cervical cancer patients.
• To assess the impact of image perturbation on feature stability.
• To compare the robustness of CT radiomics versus dosiomics features.

Main Methods:

• Retrospective analysis of 304 cervical cancer patients' planning CT images and dose maps.
• Application of random translation, rotation, and contour randomization to images and dose maps.
• Assessment of feature repeatability using intra-class correlation coefficient (ICC) and Pearson correlation coefficient (r).

Main Results:

• Radiomics features demonstrated higher repeatability compared to dosiomics features.
• Dosiomics features showed lower repeatability, particularly after small-sigma Laplacian-of-Gaussian (LoG) and wavelet filtering.
• Features extracted from original, large-sigma LoG filtered, and specific wavelet filtered images exhibited higher repeatability (ICC > 0.9).
• Positive correlations were observed between image entropy and the number of highly repeatable features.

Conclusions:

• CT radiomics features are more repeatable and robust for quantitative imaging analysis in cervical cancer.
• Dosiomics approaches require further refinement to improve feature repeatability.
• Findings support the clinical utility of radiomics for cervical cancer patient management.