Comparison of three freeware software packages for ¹⁸F-FDG PET texture feature calculation.

Area of Science:

• Radiology and Imaging
• Medical Physics
• Oncology

Background:

• ¹⁸F-FDG-PET imaging is vital for head and neck cancer assessment.
• Texture analysis quantifies intratumoral heterogeneity from PET images.
• Standardization of texture feature extraction across different software is lacking.

Purpose of the Study:

• To compare texture feature estimates derived from ¹⁸F-FDG-PET images using three distinct software packages.
• To assess the agreement and correlation of texture features extracted by CGITA, LIFEx, and Metavol.

Main Methods:

• ¹⁸F-FDG-PET images from 15 head and neck cancer patients were analyzed.
• Thirty-eight texture features were extracted using CGITA, LIFEx, and Metavol freeware.
• Statistical agreement was evaluated using Kruskal-Wallis and Dunn tests; correlation via Spearman coefficient.

Main Results:

• Significant agreement (P < 0.05) was observed for 23 of 38 texture features across all software.
• LIFEx and Metavol showed better agreement (36/38 features) compared to CGITA.
• All features were highly correlated between LIFEx and Metavol (ρ ≥ 0.70).

Conclusions:

• Texture feature extraction from ¹⁸F-FDG-PET images exhibits discrepancies across different software packages.
• These findings highlight the necessity for continued standardization efforts in radiomics.
• Establishing a reference dataset is essential for robust comparisons and clinical translation.