Enhancing radiomic feature selection in PET-based heterogeneity analysis through tumor-volume-weighted normalization

Area of Science:

• Radiomics and medical imaging analysis.
• Quantitative tumor characterization.

Background:

• Intratumoral heterogeneity assessment is crucial for tumor characterization and monitoring.
• Tumor size variations can introduce biases, making it difficult to distinguish true heterogeneity from volume effects.

Purpose of the Study:

• To develop and evaluate Tumor-Volume-Weighted Normalization (TVWN), a framework to minimize size-related biases in radiomic features.
• To preserve intrinsic image heterogeneity while correcting for volume dependence.

Main Methods:

• TVWN was tested on simulated tumors, phantoms, and longitudinal PET scans of mice.
• 97 radiomic features, including texture metrics, were analyzed before and after TVWN normalization.
• Volume dependence of radiomic features was assessed using statistical correlation.

Main Results:

• TVWN effectively reduced the influence of tumor volume across diverse conditions and feature types.
• Volume dependence of Glcm_IDN was significantly reduced (Spearman ρ: 0.290 to 0.007).
• In phantom studies, TVWN decreased the coefficient of variation for Glrlm_RE (17.75% to 3.18%) and enhanced heterogeneity discriminability (p=0.0286).

Conclusions:

• TVWN provides a scalable method for normalizing radiomic features against tumor volume effects.
• The approach improves the reproducibility and biological interpretability of radiomic data in longitudinal studies.