Tumor Volume-Aware Normalization of Radiomic Features for Robust PET Heterogeneity Quantification
Description
This dataset contains the implementation of Tumor Volume-Aware Normalization (TVAN), a multivariate radiomic normalization framework designed to remove tumor volume–dependent effects from PET radiomic features while preserving biologically meaningful heterogeneity information. The dataset includes both simulated heterogeneous tumor datasets and longitudinal preclinical PET imaging datasets derived from MyC-CAP prostate tumor-bearing mice imaged across four time points (T1–T4). The repository contains: (i) the original radiomic feature datasets, (ii) TVAN-corrected radiomic feature datasets, (iii) feature-wise tumor volume sensitivity coefficients (β), (iv) feature-ranking outputs integrating volume independence and biology preservation metrics, and (v) the Python implementation of the TVAN framework. TVAN operates by estimating feature-specific tumor volume sensitivity coefficients using regression in log-transformed feature space and removing the volume-associated component from each radiomic feature representation. The framework is intended to improve robustness, interpretability, and longitudinal consistency of PET radiomic biomarkers by minimizing size-driven confounding effects.