Evaluating patient-specific z-score images for lesion detection in low count studies
Despite excellent sensitivity, small target sizes combined with low resolution can lead to weak signals in nuclear medicine imaging. When the signal in a lesion is of the same order of magnitude as the background signal in the blood, it can be difficult to differentiate between lesion and noise. This work evaluates the concept of using patient-specific image noise information to compute a Z-score image to test for lesions with significant signal increases. Image noise is directly evaluated for inclusion in this metric using projection data resampling techniques to obtain replicate images. This approach is evaluated in the context of apoptosis imaging with technetium-99m wherein the signal in arterial voxels is measured relative to the background signal in the blood pool of the superior vena cava. In this preliminary study, the Z-score technique identifies small lesions with greater accuracy than the conventionally used target-to-background ratio (TBR) method.
|Conference||2018 IEEE Nuclear Science Symposium and Medical Imaging Conference, NSS/MIC 2018|
Cuddy-Walsh, S.G. (Sarah G.), Massalha, S. (Samia), Marvin, B.F. (Brian F.), Carey, C. (Clare), Ruddy, T.D. (Terrence D.), & Glenn Wells, R. (2018). Evaluating patient-specific z-score images for lesion detection in low count studies. In 2018 IEEE Nuclear Science Symposium and Medical Imaging Conference, NSS/MIC 2018 - Proceedings. doi:10.1109/NSSMIC.2018.8824686