Parallel-beam tilted-head analytic SPECT reconstruction: derivation and comparison with OSEM

Parallel-beam tilted-head single photon emission computed tomography (TH-SPECT) was previously implemented on a SPECT system for its potential to image breast lesions and nearby axilla of seated, upright women. All TH-SPECT reconstructions will contain artifacts since the tilted orbit does not satisfy the Orlov sampling criteria. However, it is not clear which reconstruction method, if any, is better suited for TH-SPECT data. Here a geometric derivation of the ramp filter for tilted parallel-beam geometries is presented. A filtered backprojection (FBP) algorithm, using this filter, was then implemented and compared with an iterative ordered subsets expectation maximization (OSEM) algorithm, using TH-SPECT data. A breast scan at various tilt angles was simulated and used to generate a noise versus bias study for both methods. Contrast and signal-to-noise ratio (SNR) values, as well as axial elongation present in all TH-SPECT reconstructions, were characterized using a mini-Defrise disk phantom placed inside a fillable breast phantom and imaged from 0 to 15/spl deg/ head tilt. A fillable breast phantom containing lesions was also imaged with a system dedicated to prone breast SPECT from 0 to 30/spl deg/ to evaluate the effects of incomplete sampling due to greater tilt angles. FBP noise versus bias studies indicated a greater increase in bias with tilt angle compared to OSEM reconstructions. At small tilt angles about the mini-Defrise disk phantom, poorer contrasts were obtained with FBP compared to OSEM at similar noise levels. All reconstructions of the fillable breast phantom indicated axial elongation at greater tilt angles, although FBP reconstructions displayed an increase in stretching distortions of the breast. OSEM SNR and contrast values were higher at all degrees of tilt. In conclusion, measured results indicate OSEM TH-SPECT reconstruction provides better contrast and SNR values and may offer better shape and uniform activity distribution of the breast compared to FBP methods.