SPECT reconstruction directly from photomultiplier tube signals.

Abstract

Compton scatter in the object is an unsolved problem in single-photon emission computed tomography (SPECT). Current correction schemes in SPECT can be divided into three categories: those correction schemes that examine the spatial contribution of scattered photons, those schemes that examine the energy signature of scattered photons, and those techniques that examine both the spatial and energy aspects of detected photons. Our approach to scatter correction most closely resembles those in the third category. The SPECT systems that have been developed at the University of Arizona (UA) are stationary systems which consist of modular gamma cameras that view the object through a multiple-pinhole coded aperture. Characteristic of the UA SPECT systems is that the system response function (position-sensitive point-spread function) is measured; measuring the response function allows us some flexibility in our choice of the detector space in which we work. In a typical scintillation-camera system, the photomultiplier-tube (PMT) response to a gamma-ray interaction is used to estimate the interaction position and energy of the absorbed gamma ray. We call this estimate the pixel-space data. We could also use the digitized PMT signals directly as our data. Using PMT signals directly we avoid the possibility of misestimating the spatial and energy coordinates of scintillation events. We have proved that reconstructions from PMT-space data are of at least comparable quality to reconstructions from data in pixel space. In order to quantitate the difference between PMT and pixel spaces, we performed simulation studies and used the signal-to-noise ratio (SNR) of the ideal observer as our image quality metric. We simulated the projection of photons scattered zero to four times, and detected the simulated flux with cameras of different energy resolution and with the modular camera. We found that for the detection of a small cold tumor in a uniform background, the energy resolution of the simulated camera changed the SNR very little and there was little difference between the SNR from data in PMT and pixel spaces.