Three-dimensional oxygen maps of tumors in real time – Analysis in the context of active tumor vasculature

Background and Objective Characterizing the tumor microenvironment (TME) requires integrating multiple physiological features, including oxygenation, vascularity, and redox status. While EPR oxygen imaging (EPROI) provides spatial pO₂ maps, conventional metrics such as median pO₂ or hypoxic fraction (HF) may not fully capture vascular characteristics. This study aimed to identify EPROI-derived parameters that best reflect tumor vasculature by correlating oxygenation metrics with Doppler ultrasound vascular imaging and redox kinetics. Methods Mel270 uveal melanoma tumors were implanted in the intrascapular fat pad of SCID mice, providing a highly vascularized niche. Non-invasive imaging included ultrasound (anatomical and Doppler) and EPROI for pO₂ mapping. EPR spectroscopy of the nitroxide redox probe was used to estimate tissue redox status. Parameters extracted included median pO₂, HF20, VC40, VC60, vascular fraction (PV), and kinetic descriptors (α, β, mean amplitude). Correlation analysis and Least Absolute Shrinkage and Selection Operator (LASSO) regression, with Leave-One-Out Cross-Validation (LOOCV) and Partial Least Squares Regression (PLSR), as well as principal component analysis (PCA) were performed to identify the most informative metrics. Results VC40 and VC60, representing the upper tail of the pO₂ histogram, showed moderate correlations with vascularity (PV), median pO₂, and redox kinetics (β), outperforming HF20. Both LASSO and PCA confirmed VC40/VC60 as the most robust single parameters for vascular characterization, as well as the best predictors of median pO2, with a secondary contribution from mean amplitude and PV. As expected, median pO₂ correlated negatively with metastasis, while HF20 correlated negatively with PV in PCA analysis. Tumor size did not correlate with oxygenation or vascularity. Conclusions Advanced evaluation of the tumor microenvironment requires complementary parameters like VC40/VC60 and HF20/pO₂ to capture its full complexity, integrating oxygenation, vascularity, and redox data. Future work should adopt advanced histogram-based and machine learning methods, as widely used in MRI, to fully exploit spatial oxygen and vascular network imaging data.