ROI value calculation methods for automated quantitation of mouse PET studies

Objectives ROI value calculation methods that could compensate for errors/imperfections in ROI definition obtained by automated image warping procedures have been investigated.

Methods Seven normal mice (C57BL/6; male; 26.0–31.5 g) and seven fasted mice (C57BL/6; male; 23.1–26.8 g) were scanned for FDG PET and CT images. Using CT image as a guide for warping to the common space, each PET study was registered to the Digimouse atlas by SPM. Digimouse ROI template was applied to the warped PET images. The following four different automated calculation methods were used to obtained the ROI values in 6 major organs (liver, right kidney, heart, lung, brain) and were compared to those based on the manually drawn ROIs. MN: mean of pixels within ROIs; MG: mean of pixels within organ boundary determined by max gradient in ROI; MH: mean ofpixels within ROI according to histogram of pixel value; MD: median of all pixel values within ROI. Last, all ROI values were converted to standard uptake values (SUV).

Results SUV of one or more of the evaluated methods on the Digimouse template of the six organs matched well with those based on manually drawn ROIs in all 14 mouse PET studies. For brain, lung, and liver, method MN had the best agreement with bias of less than 7%, and r-square between 0.84 and 0.94; for heart, kidney, and bladder, method MG had the best agreement with bias of less than 6% and r-square between 0.94 and 0.99.

Conclusions Coupling of appropriate ROI calculation methods with Digimouse ROI templates and mouse PET images warped to the common space (Digimouse) can ensure reliability of the ROI value determinations and will allow the quantitative analysis of mouse PET images to be automated.