Experience with 16-module QUAD-Hidac
John Missimer
Paul Scherrer Institute
05/04/02
Physical characteristics
The 16-module variant of the 3D Quad-Hidac camera [1] consists of four detector banks of four planar high density avalanche chambers each. The banks are arranged to provide an active imaging volume of 170mm x 170mm x 280mm, which implies a solid angle cov-erage of 75% for a point source at the center; the currently supplied filtered backprojec-tion algorithm [2] accommodates a reconstructed field of view (FOV) of 15cm in diameter and 25cm in axial extent. Each detector module consists of a multiwire proportional chamber combined with laminated plates of interleaved lead and insulating sheets mechanically drilled with a dense matrix of small holes. The holes are .4mm in diameter and .5mm from center to center, yielding intrinsic submillimeter resolution. In order to assure uniform response, the detector banks rotate continuously 180 deg back and forth in a 6 second cycle. Data are acquired in list mode, implying that time frames, reconstructed FOV and voxel sizes can be defined at reconstruction. Only one acquisition is required for the full FOV. The voxel size can be chosen between 0.1mm and 1mm for cubic voxels.
Stability
Delivered to PSI in mid-December 2000, the camera has undergone daily performance checks using a commercial point source of 22 Na have been made of the camera since early March 2001. The camera is calibrated daily. A fit to the time course of the total coin-cidence rates yields a raw detection efficiency of 1.5%. Deviations from the fitted decay curve do not exceed 0.25% of the total coincidence rate.
Linearity
The region of linear count rate extends up to about 220000 cps with a 18 F-line source of about 16 MBq. For a phantom of volume 388ml, outside diameter 5cm and length 25cm, which simulates a rat, we expect linearity up to about 30 MBq. The rates can be multi-plied by factor of two for the 32-module version.
Resolution
Images of the point source were reconstructed using filtered back projection [2] with voxel size 0.1mm. The measurements yielded: radial 1.08+-0.03 FWHM (2.07+-0.06 FWTM), tangential 1.11+-0.05 (2.21+-0.10) and axial 1.05+-0.03 FWHM (2.06+-0.09). Deviations on the axis from the center to 40 mm were not significant. With the resolution recovery incorporated in the list mode reconstruction program developed by Andrew Reader[3], the resolution reduces to about 0.7mm in all three components.
Dead time
Measurements with the 18 F-line source indicate an almost linear decrease in detection efficiency up to a coincident rate of 220000 cps. The total decrease was about 15%.
Randoms
The rate of random coincidences, computed from the singles rate, reaches 30% of the total coincidences for the 18 F-line source with 16 MBq activity. Preliminary measure-ments using a cylindrical phantom with the approximate dimensions of a rat: 5cm in diameter and 25cm in length having 388ml volume, indicate that the randoms constitute about 50% of the total rate for 18 MBq of 18 F. We are in the process of evaluating the spatial variation and object dependence of the random rates.
Scatter
Scatter, occurring between the modules, determined by a threshold technique applied to the reconstructed images of the point source, accounted for 29% of the acquired events. The determination of scattered coincidences is a considerable overestimate due to the 1.2 MeV gamma emitted by 22 Na. Thus, the net efficiency for true coincidences exceeds 1%, or, equivalently, a point source absolute sensitivity of 10 cps/kBq. We are in the pro-cess of evaluating the spatial variation and object dependence of the scatter.
Absorption
For the cylindrical phantom, total attenuation is about 35%. We are analysing the spatial variation presently; my impression is that it is slight. No transmission source is supplied with the camera, nor is any offered. With the number of lines of response which are typically acquired, a transmission scan would hardly be practical, and, in my opinion, the best solution would be to use attenuation correction computed from the emission scan.
Noise equivalent count rate and calibration factor
This is work in progress.
Computer and Operating System
The camera was delivered with a high performance, dual processor Dell PC operating under NT. The processors of the currently offered system feature 2 Ghz and 2x2 GByte RAM, which permit simultaneous acquisition and reconstruction.
Software
Executables to manage acquisition and perform reconstruction are provided with the camera. They include a graphic users interface to set acquisition and reconstruction parameters and offer basic viewing and analysis of the acquisition in progress and of the reconstructed images. The file formats are documented; the default format is a header followed by binary data. There is an option to convert the reconstructed images to ANALYZE. Andrew Reader[3] of Manchester, England intends to release his list mode statistical reconstruction program as open source. We routinely use a preliminary version for reconstruction. Andrew s code offers options for introducing corrections, and he is also working on attenuation and scatter correction.
Reliability and Service
The camera is very reliable. It has been operating since we installed it with only one interruption due to a problem in the gas system which may have resulted from a defective argon bottle. We had one defective high voltage module on delivery. Such defects reduce the sensitivity by about 10%, but cause no spatial distortion because the modules rotate around the full field of view. OPS has been very helpful in isolating and repairing the problems. Besides servicing vis-its, we ve received quite a bit of advice via telephone and email, and were able to do some repairs ourselves with spare parts prepared and sent to us. Like any new system, it takes some time to become acquainted with and acquire confidence in it, but the camera is certainly robust enough and well designed that its operation quickly becomes routine. I think OPS has gained a lot of useful experience with us, too.
Acquisitions
Acquisitions of about 50 rats and mice have been performed with F, C and Cu tracers since we obtained a license for animal experiments in June 2001. The acquisition times varied between 30 minutes and 2 hours, and provided data for 12 to 20 1 to10 min time frames. Recently high resolution images of mouse striata have been acquired, and are presently being analyzed.
Some images acquired at PSI: image archive.
References
[1] A P Jeavons, R A Chandler, C A R Dettmar. "A 3D HIDAC-PET Camera with Sub-mil-limetre Resolution for Imaging Small Animals". IEEE Trans. Nucl. Sci. 46 (June 1999) 468-473
[2] P Kinahan and J Rodgers, "Analytic 3D Image Reconstruction using All Detected Events", IEEE Trans. on Nucl. Sci., NS-36 (1989) 964-968
[3] AJ Reader et al., Regularized one-pass list-mode EM algorithm for high resolution 3D PET image reconstruction into large arrays . Proc. IEEE Medical Imaging Confer-ence, San Diego (2001)