CzT camera
All tests were performed on the first VERITON camera (Spectrum Dynamics, Caesarea, Israel) commercially available (FDA and EU clearance was obtained for the present study), installed at the University Hospital of Caen, France. The camera consisted of 12 columns of detectors arranged in a ring configuration. Each column was composed of a 16 × 128 array of CzT pixel units and equipped with high sensitivity tungsten parallel-hole collimators in alignment with the pixel array. Detector surfaces were equipped with skin sensors to get as close to the body contours as possible during acquisition. Acquisition could be performed either in manual mode with a circular field of view or in body-contouring mode. Acquisition consisted of a few angular acquisition steps ensured by a gantry rotation. The number of required steps, typically four for normal body contours, was dependent upon the size of the field of view. At each acquisition step, a sweeping motion of each column of detectors was performed to collect data from the entire field of view. A focus mode was also available to force detector movement to collect data in a region of interest defined by the user on a pre-scan acquisition. Figure 1 describes the general camera architecture and explains the focus mode. In this study, all acquisitions were performed with a default energy window set by manufacturer at 20% centred on the photopeak of 99mTc and a matrix size of 256 × 256, resulting in a squared pixel size of 2.46 mm. Images were reconstructed with a proprietary OSEM algorithm including the use of an internal pre-iteration median convolution kernel for noise suppression, with 10 iterations, 32 subsets and without the use of a post-reconstruction filter.
Anger camera
The Anger camera used in this study was a dual-head Symbia camera (Siemens, Erlangen, Germany) equipped with a 3/8 inch NaI (Tl) crystal. Detectors were equipped with a low-energy, high-resolution (LEHR) collimator. A full calibration of the detector was done before the evaluation, and the camera passed all calibration tests. SPECT acquisitions were performed with standard reconstruction parameters used in clinical routine as recommended by the manufacturer. With the exception of some tests that needed default parameters to be adjusted, all acquisitions were performed with the following parameters: an energy window of 15% centred at 99mTc photopeak (140 keV), 64 projections with a matrix size of 128x128, resulting in a 4.8-mm-square pixel size. Reconstructions were performed with a 2D-OSEM algorithm with 8 iterations and 4 subsets and without post-filtering.
Patient examinations
As part of the FDA clearance scheme, several patients were imaged within the framework of a phase I study approved by the regional Ethics committee under the number ID RCB: 2017-A01448-45, registered at clinicaltrials.gov with the following number: NCT03438123. We selected cases representative of various isotopes, organ sizes and shapes, as well as count statistics.
Extrinsic resolution
Extrinsic resolution was assessed using 1-mm-diameter capillaries (PTW Freiburg) placed in air and filled with approximately 20 MBq/mL of 99mTc. The three-line sources were positioned parallel to the longitudinal axis of rotation, 0, 4.5 and 9 cm from the centre of the field of view. Five-minute tomographic acquisitions were performed on both cameras with a circular shape, a radius of rotation of 15 cm and a matrix size 256 × 256 to obtain the finest linear sampling. The resulting pixel size was 2.46 mm for the CzT camera and 2.40 mm for the Anger camera. Acquisitions were performed on the Anger camera with LEHR collimators. Different sets of images were reconstructed using an OSEM reconstruction algorithm by varying the total number of iterations from 8 to 640, except for the Anger camera for which the maximum number of iterations was limited to 320, and without post-reconstruction filters applied. Profiles passing through the maximum pixel value of each source were drawn in radial and tangential directions. A Gaussian function was fitted to compute the full width at half maximum (FWHM) of each profile measurement.
Tomographic resolution
As final image resolution depends on both acquisition and reconstruction parameters, tomographic resolution was estimated with phantoms simulating different clinical situations used in routine.
To mimic a clinical brain acquisition, tomographic resolution was first estimated using a head phantom (from International Electrotechnical Commission (IEC) Standard 61675-2) equipped with three cavities located in the middle and at 4.5 and 9 cm from the centre. The same line sources used for extrinsic resolution measurement were introduced in the phantom filled with a background activity of 56 MBq of 99mTc. The phantom was positioned at the centre of the camera, and acquisitions were performed with a circular radius of rotation of 15 cm. For the CzT camera, a 640-s acquisition was performed with a matrix size of 256 × 256, and for the anger camera, 64 projections of 10 s per head were performed with a matrix size of 128 × 128 and a zoom factor 1.45.
To mimic a clinical body acquisition, the same experiment was then carried out using a body phantom (from IEC standard 61675-1) filled with a 97-MBq background activity of 99mTc using the previous line sources placed at the same distance from the centre of the phantom. Acquisitions were performed on both cameras with the phantom placed at the centre of field of view in autocontour mode. For the CzT camera, a 640 s was performed with a matrix size of 256 × 256, and for the anger camera, 32 projections per head of 20 s were performed with a matrix size of 128 × 128 without zoom factor. All acquisitions were reconstructed by varying the total number of iterations from 8 to 640, except for the Anger camera for which the matrix size used limited the maximum number of iterations to 320 for the head phantom and 160 for the body phantom, and by applying a Gaussian post-reconstruction filter of 5-mm FHHM. A profile passing through the maximum pixel value of the sources was drawn. A Gaussian function was fitted to determine the FWHM of each line source in the radial and tangential directions.
Energy resolution
Energy resolution was measured for the most commonly used radioelements in nuclear medicine: 201Tl (70 keV, 167 keV), 99mTc (140 keV), 123I (159 keV) and 111In (171 keV, 245 keV). Source preparation and spectral acquisition were carried out according to manufacturer’s recommendations. For the CzT camera, energy spectra were acquired using fillable cylinders of around 42 cm-length and 0.9 cm of inner diameter, filled with a solution of 110 to 370 MBq depending on the tested radioisotope and placed at the centre of the field of view. On the Anger camera, point sources were used instead with an activity adjusted so that the count rate measured on the camera was between 15 and 50 kilocounts per second at acquisition start and centred in the field of view equidistant from the 2 detectors equipped with LEHR collimators. Energy spectra were acquired for all pixel elements on the CzT camera and for the two detectors on the Anger camera. Spectral acquisition was stopped when the number of counts per pixel reached 25 kilocounts for the CzT camera, and when the number of counts recorded at the maximum energy peak was about 32 kilocounts for the Anger camera. Each peak of energy was fitted by a Gaussian function to calculate FWHM. The resulting energy resolution was expressed as the mean FWHM for all pixel units +/− Standard Deviation (SD) for the CzT camera and as the mean FWHM for the two detectors +/− SD for the Anger camera.
Tomographic sensitivity
The purpose of this test was to assess tomographic sensitivity under conditions used in clinical routine rather than to evaluate the intrinsic sensitivity of the system. Therefore, the default parameters used in clinical routine were selected. As the global sensitivity of the system depends both on the nature of the detectors and on the geometry of detection, sensitivity was assessed for different types of radiation sources, including conditions representative of clinical situations. Sensitivity was first evaluated on each camera with a point source of approximately 25 MBq of 99mTc placed at the centre of the field of view and a radius of rotation of 15 cm. Measures were successively performed with the source placed in air, at the centre of a water-filled head phantom and at the centre of a water-filled body phantom (phantoms are those described in the tomographic resolution section). Two different count measurements were performed on the CzT camera: one on the entire field of view and a second with a focus of the detector on a circular region of interest centred on the point source. Sensitivity was then evaluated with the head and the body phantoms filled with a uniform solution of approximately 350 MBq of 99mTc and placed at the centre of the field of view. Acquisition was performed with a radius of rotation of 15 cm for the head phantom and in autocontour mode for the body phantom. The exact activity of the prepared sources was assessed by applying radioactive decay correction to the measurement performed with a calibrated well counter (MEDI-405, Medisystem, Guyancourt, France). The total number of counts was recorded for each acquisition of 450-s duration, and the sensitivity was calculated as follows:
$$ \mathrm{Sensitivity}\kern0.35em \left(\mathrm{counts}.{\mathrm{MBq}}^{-1}.{s}^{-1}\right)=\frac{\mathrm{total}\kern0.35em \mathrm{recording}\kern0.35em \mathrm{counts}}{\mathrm{acquisition}\kern0.35em \mathrm{time}\ (s)\times \mathrm{source}\kern0.35em \mathrm{activity}\ \left(\mathrm{MBq}\right)} $$
Image quality
A Flangeless Deluxe Jaszczak phantom filled with 350 MBq of 99mTc was scanned on each camera to evaluate global image quality, lesion detectability and uniformity of the reconstructed slice. Phantom was positioned on a headrest at the centre of the field of view, and acquisitions were performed in autocontour mode with the default parameters described above. Focus mode was not activated on CzT camera. Images were acquired on Anger camera with a zoom factor of 1.45 resulting in a squared pixel size of 3.3 mm. To evaluate the influence of acquisition time, four acquisitions were tested as follows: 30, 20, 10 and 5 min. Images were corrected for attenuation with the Chang attenuation method, applying a coefficient of attenuation of 0.11 cm-1. For each acquisition, the total number of counts was recorded. To easily compare the SPECT performance of the two cameras, image quality was quantitatively assessed by computing contrast, uniformity and noise index values, as proposed by the American Association of Physicists in Medicine [12]. All regions of interest and statistical tests were computed using the AMIDE software [13].
Image contrast
To calculate the image contrast for each dataset, the transverse slice where the cold spheres were the most visible was selected. The minimum pixel counts were determined for each spheres in the chosen slice, and the mean pixel value was calculated in a 15 × 15-pixel square ROI drawn in the uniform background of the phantom. The image contrast was then calculated for each cold sphere as follows:
$$ \mathrm{Contrast}=\frac{\mathrm{mean}\kern0.35em {\mathrm{pixel}}_{\mathrm{background}}-\min \kern0.35em {\mathrm{pixel}}_{\mathrm{coldsphere}}}{\mathrm{mean}\kern0.35em {\mathrm{pixel}}_{\mathrm{background}}} $$
Tomographic uniformity
To calculate the image uniformity, a 15 × 15-pixel square ROI was drawn on a transaxial slice located on the uniform part of the phantom. The mean, maximum, minimum and standard deviations of the pixel values within the ROI were recorded. Integral uniformity and root mean square noise (RMS noise) were calculated with the following formulas:
$$ \mathrm{Integral}\kern0.35em \mathrm{uniformity}\kern0.35em \left(\%\right)=\frac{\left(\max \kern0.35em \mathrm{pixel}-\min \kern0.35em \mathrm{pixel}\right)}{\max \kern0.35em \mathrm{pixel}+\min \kern0.35em \mathrm{pixel}}\times 100 $$
$$ \mathrm{RMS}\kern0.35em \mathrm{noise}\left(\%\right)=\frac{\mathrm{standard}\kern0.35em \mathrm{deviation}}{\mathrm{mean}\kern0.35em \mathrm{pixel}\kern0.35em \mathrm{value}}\times 100 $$
Clinical images
Four clinical cases were studied to compare image quality obtained on the CzT and Anger cameras: a 99mTc-pertechnetate thyroid scan (15 min after injection of 117 MBq, 10 min scan duration), a 99mTc-HMDP bone scintigraphy (3 h after injection of 643 MBq, 10-min scan duration), a 99mTc-DMSA renal scan (4-h after injection of 111 MBq, 10-min scan duration ) and a prostate cancer treatment imaging study after injection of 223Ra (48-h after injection of 5.27 MBq, 30-min scan duration). Patients were scanned successively on each camera with the same acquisition time. The patients were first scanned on the Anger camera and then on the CzT camera with correction of scan duration for activity decay. To quantitatively assess image quality, the contrast-to-noise ratio (CNR) was computed for each lesion detected. Two spherical volumes of interest (VOI) of 5-mm diameter were manually drawn on clinical images using the AMIDE software [13]: the first one centred on the lesion and the second on healthy tissue near the lesion. CNR was calculated as follows:
$$ \mathrm{CNR}=\frac{\mathrm{mean}\ {\mathrm{pixel}\ \mathrm{value}}_{\mathrm{lesion}}-\mathrm{mean}\ {\mathrm{pixel}\ \mathrm{value}}_{\mathrm{healthy}\ \mathrm{tissue}}\ }{{\mathrm{standard}\ \mathrm{deviation}}_{\mathrm{healthy}\ \mathrm{tissue}}} $$
All images were analysed, and VOIs were manually drawn jointly by a nuclear medicine physician and an expert in medical physics with more than 10 years of experience.