Targeted reconstruction for the short-duration SPECT imaging: Simulation studies

Objectives Our aim is to develop a SPECT reconstruction algorithm which would allow us to decrease the duration of clinical studies where small organs or tumors are surrounded by a relatively uniform low-activity background.

Methods Our targeted reconstruction (method M2) relies on the extraction of the region of interest (ROI) surrounding an object (organ or tumor) from the conventionally reconstructed (by method M1) SPECT image. After the rough delineation of ROI, it contains some portion of the surrounding medium and it’s size exceeds the size of the object ~10 times. The activity is reconstructed only inside ROI using MLEM algorithm where the estimated noise-free distribution in non-ROI media is added into the denominator. In this study, we address the question if this targeted (performed only for ROI with considerably fewer voxels than in M1) reconstruction may allow us to reduce the required number of camera stops. Two simulations imitating thyroid SPECT studies with the organ/background ratios 1:10 and 1:20 were performed. The images from 60 (as in clinical protocols), 20, 10, and 6 projections were reconstructed using conventional and targeted MLEM methods with up to 60 iterations. While the conventional algorithm M1 reconstructed activity in 128x128x128 voxels, the targeted one dealt only with 7411 voxels.

Results While methods M1 and M2 resulted in similar bias (difference between true and calculated activities in object) for 60 and 20 projections, method M2 prevailed over M1 for 20 and 6 projections. In the last situation, the bias after 60 iterations was equal to 35% for M1 and 25% for M2. However, the variance measured as a relative standard deviation in the background (inside ROI) increased more rapidly for M2 than for M1. The bias/variance analysis showed advantages of targeted method which provided lower variance for the majority of biases.

Conclusions The targeted reconstruction algorithms are less sensitive to the decrease of the number of camera stops than the conventional methods and provide equivalent images from fewer projections