TH‐C‐WAB‐05: Blind Functional Volume Segmentation in PET

Abstract

Purpose: To propose a blind PET functional volume segmentation method that does not need any prior calibrations. Methods: The partial volume effect (PVE) in PET imaging was considered as a forward problem that convolves the original target object with the point spread function (PSF) of the PET scanner, and made a PET image blurred. We showed that the optimal threshold for accurate object localization in a PET image is a function of the ratio of the object size to the PSF standard variance. We defined a new quantity, the optimal volume ratio (OAR), which represents the ratio of the segmented volume when using the optimal threshold to the whole target volume with greater activity than the background. The OAR can be expressed as a function of the optimal threshold, and was then used to derive the optimal threshold through a sequence of region growing algorithms with different thresholds. The new method avoids solving the ill‐conditioned inverse problem, and is fully blind to the PET scanner. The proposed method was tested on PET phantom with known object location and extension, and its robustness with respect to source‐to‐background ratio (SBR) and object size was evaluated. Results: At all SBR levels and for all object sizes, the new blind segment method performed similarly to existing methods with prior calibrations, and outperformed existing methods that do not take into account PVE. In addition, the new blind segment method is robust to the changes of SBR level and object size. Conclusion: The results showed that the proposed blind method did not need any prior calibrations and performed well for the functional volume segmentation in PET.