Target localization using scanner-acquired SPECT data.

Abstract

Target localization using single photon emission computed tomography (SPECT) and planar imaging is being investigated for guiding radiation therapy delivery. Previous studies on SPECT-based localization have used computer-simulated or hybrid images with simulated tumors embedded in disease-free patient images where the tumor position is known and localization can be calculated directly. In the current study, localization was studied using scanner-acquired images. Five fillable spheres were placed in a whole body phantom. Sphere-to-background 99mTc radioactivity was 6:1. Ten independent SPECT scans were acquired with a Trionix Triad scanner using three detector trajectories: left lateral 180°, 360°, and right lateral 180°. Scan time was equivalent to 4.5 min. Images were reconstructed with and without attenuation correction. True target locations were estimated from 12 hr SPECT and CT images. From the 12 hr SPECT scan, 45 sets of orthogonal planar images were used to assess target localization; total acquisition time per set was equivalent to 4.5min. A numerical observer localized the center of the targets in the 4.5 min SPECT and planar images. SPECT-based localization errors were compared for the different detector trajectories. Across the four peripheral spheres, and using optimal iteration numbers and postreconstruction smoothing, means and standard deviations in localization errors were 0.90 ± 0.25 mm for proximal 180° trajectories, 1.31 ± 0.51 mm for 360° orbits, and 3.93 ± 1.48 mm for distal 180° trajectories. This rank order in localization performance is predicted by target attenuation and distance from the target to the collimator. For the targets with mean localization errors < 2 mm, attenuation correction reduced localization errors by 0.15 mm on average. The improvement from attenuation correction was 1.0 mm on average for the more poorly localized targets. Attenuation correction typically reduced localization errors, but for well-localized targets, the detector trajectory generally had a larger effect. Localization performance was found to be robust to iteration number and smoothing. Localization was generally worse using planar images as compared with proximal 180° and 360° SPECT scans. Using a proximal detector trajectory and attenuation correction, localization errors were within 2 mm for the three superficial targets, thus supporting the current role in biopsy and surgery, and demonstrating the potential for SPECT imaging inside radiation therapy treatment rooms.