Patient-Informed Lesion Insertion Framework for Realistic Quantitative Evaluation In PET Imaging
Abstract
Purpose
Accurate quantitative assessment in positron emission tomography (PET) is essential for reliable lesion characterization and therapy response evaluation. Quantitative accuracy is conventionally investigated using physical phantoms or numerical simulations; however, these approaches rely on simplified assumptions regarding lesion geometry and background activity, limiting their ability to reflect true clinical conditions.
Methods
To address this, we developed an open-source Python framework for controlled lesion insertion in clinical PET studies. Built on the PyTomography engine, the framework inserts either idealized spheres or realistic lesions directly into real patient PET raw data. These realistic lesions were derived from a dataset consisting of 1579 lymphoma lesions (from 76 PET patients), spanning a wide range of anatomical sites, lesion sizes, and uptake patterns. The workflow employs statistical ablation to suppress background activity at the insertion site, forward-projects the defined lesion, and merges it with the original raw data for reconstruction using various iterative algorithms. We utilized the proposed framework to compute recovery coefficients as a practical estimator of partial volume effects (PVE). Spherical and real lesions were inserted into bone, lung, and liver at multiple sizes per site. RCs were calculated to quantify activity recovery.
Results
The results showed that PVE depends not only on lesion size but also strongly on lesion morphology, compactness, and background region. At 2000 mm³, real lesions showed 0.18 lower RC than spheres. This bias persisted at 4000 mm³ (liver: 0.71 vs 0.80), confirming systematic RC overestimation by spherical models.
Conclusion
This work reframes quantitative PET evaluation from a phantom-driven exercise to a patient-informed paradigm. This approach enables more meaningful interpretation of quantitative metrics and provides a scalable pathway toward patient-relevant validation, optimization, and harmonization of PET reconstruction strategies. The lesion library will be shared with the community as part of the framework.