General device features
The device relies on gamma ray detection to determine the amount of activity in the blood. The system was designed to accommodate two detector modules, facing each other as shown in Fig. 1. These modules can be operated independently with the objective of increasing the sensitivity or in coincidence mode to reduce background counts. For the prototype evaluation, a single detector module was used, and therefore, coincidence counting was not performed. The prototype dimensions are 36 × 29 × 15 cm3 including all components: power supplies, electronic boards, motors and pump. The blood is withdrawn from the patient via a catheter (arterial or venous) connected to a peristaltic pump. A pump was dedicated to laboratory testing (model 313VDL, Watson Marlow, Falmouth, UK), while a second one, achieving lower withdrawal rates, was used in the patient study (model P625, Instech Laboratories, Plymouth Meeting PA, USA). The catheter delivers blood to the gamma counting system and then to a waste container. Figure 1 shows the overall organisation of components of the device. The blood withdrawal rate depends on the catheter size and can be set between 3 mL/min to 10 mL/min for the 313VDL pump and between 1 mL/min to 7 mL/min for the P625 pump.
The total amount of blood withdrawn from a patient should be kept as low as possible. McGuill et al. suggested to withdraw no more than 7.5% of the total blood volume [16]. It is therefore important to adjust the withdrawal parameters to the pharmacokinetic behaviour of a given radiopharmaceutical. The device built allows pre-programmed, variable pump rates in order to minimise the total amount of blood withdrawn for a given study while capturing the dynamics of uptake. Everett et al. have reported that in a 924 patients PET study, taking 117 to 137 mL of blood by arterial cannulation led to a single case of adverse effect (thrombotic occlusion) and concluded that the practise was safe, even if the catheter was in place for 5 h on average [17]. Zanotti-Fregonara et al. came to the same conclusions from their experience with more than 3000 patients [1].
The device must be located as close to the patient as possible to reduce activity dispersion along the catheter. Therefore, a design as compact as possible was sought for the device. The weight of the prototype was approximately 10 kg, including shielding. This allows the use of the device directly on the patient bed, thereby minimising the length of catheters required and reducing the effect of diffusion that affects the time resolution. The activity dispersion along the catheter can be estimated and corrected, typically by monoexponential deconvolution [18, 19] or step function calibration [20], but this was not implemented yet for the prototype device presented here. An access to radial blood with short catheters is more complicated in the case of brain studies where arms usually rest alongside the patient. In these cases, the system can be positioned at the feet of the patient with a longer catheter, at the expense of larger dispersion. For head studies with one or two bed positions, another possibility is to place the device on a cart beside the patient with a shorter access to the arm.
Gamma detector
In order to fulfill compactness requirements, a CdZnTe (CZT) semi-conductor detector was selected. A 20 × 20× 15 mm 3 commercially available CZT crystal from Redlen Technologies (Saanichton BC, Canada) was chosen for the prototype, primarily for its large volume. Although this detector was designed primarily for imaging applications [21, 22] and has an 11 × 11 anode readout scheme, it provides the large detection volume required for the counting application developed here. Pixels on the detector are 1.22 mm in size deposited at a pitch of 1.72 mm. A grid of 0.1 mm is deposited 0.2 mm around the pixels, except at the edges of the pattern where it is 0.5 mm wide. For counting-only purposes, the readout pins of the 121 pixels were connected to obtain a pattern similar to a coplanar grid [23]. A custom-made front-end electronic on a six-layer PCB was used to create a virtual coplanar detector where pixels are connected column-wise, leading to a two-channel readout scheme where five columns are interleaved between six others and can be maintained at two different biases [15]. Figure 2 shows one layer of the PCB and conductive tracks used to make a virtual coplanar detector from a pixelated detector. This anode geometry with alterning columns maintained at different biases has the advantage of preserving energy resolution comparatively to a planar geometry and requires only two polarised anodes for readout [24, 25]. A custom-made charge sensitive preamplifier for each anode allowed the creation of a very compact board that also includes the appropriate routing of pixels to create a coplanar readout. The output of the preamplifier was fed to a dual-channel ultralow noise amplifier (AD8432, Analog Devices, MA, USA) which was used to obtain a differential signal that was then routed to the device’s main board by a Mini DisplayPort cable. The CZT crystal/preamplifier assembly was packaged in a compact custom-made 27 × 67 × 37 mm3 aluminium casing as shown in Fig. 2, along with 3D-printed pieces for accurate and reproducible positioning of all components. The weight of the detector module is 146 g.
The detector assembly was shielded with 25 to 35 mm of 97% pure tungsten. A custom-made plastic container was 3D-printed and filled with tungsten cubes, as shown in Fig. 3. The container, which has a slit for the catheter, ensures reproducible positioning. A separate 3D-printed piece (not shown) slides in the slit and maintains the catheter in place. The length of the catheter exposed to the CZT detector is 29 mm. The detector was not shielded where the cables pass; this area was pointing towards the ceiling in the clinical experiments as shown in Fig. 8. In all cases, there was no direct unshielded line of sight between the CZT crystal and the main source of background radiation (the patient).
Data acquisition
A field-programmable gate array (FPGA)-based (Cyclone V, Altera, CA, USA) circuit was designed to control the acquisition and the different subsystems of the blood counter. The FPGA chip allows a convenient handling of signals and components through the Nios II processor. The signal workflow is illustrated in Fig. 4. The analog signal from the detector-preamplifier assembly is digitised by a “free-running” quad, 14-bit analog-to-digital converter (ADC, AD9253, Analog Devices, MA, USA) and then routed to the FPGA where programmable filters and thresholds (shaping) are applied. More specifically, the two signals from the virtual coplanar detector are combined (weighted subtraction) on the FPGA to compensate for charge trapping, as described by Luke et al. [23, 26]. A thorough characterisation of the detector, not reported here, allowed to set optimal operation parameters of this anode configuration. A trapezoidal filter was applied on the resulting signal, and the maximum was extracted. Counts exceeding 70 keV were stored in memory while the FPGA is waiting for calls by the host program via a USB connection.
Graphical user interface
The host program runs on a PC and has a graphical user interface (GUI) implemented with Qt (version 5.3, Helsinki, Finland). The GUI, shown in Fig. 5, can display the detected counts per second for two energy windows and two detector modules. The centre panel gathers the primary controls that allow an acquisition to start, stop and reinitialise. The information on the acquisition sequence is also shown in the central panel. It defines the pump rates and acquisition duration. Sequences can be programmed, saved and reused. The GUI also shows a visual representation of a carousel where discrete blood samples can be packaged in evacuated tubes for further analysis. For this work, the packaging feature of the device was not used.
The bottom part of the GUI shows either a graph of the activity as a function of time or an energy histogram of detected events, depending on which tab is selected. Four markers can be positioned to define two energy windows, allowing for example dual-isotope studies. The energy windows selected are applied to the activity displayed in the associated tab.
The GUI has a user mode for regular use and a superuser mode for development. The superuser mode allows a manual control of the device and real-time programming of pump rates, motors and data processing parameters. The user mode is meant to be used in a clinical setting and therefore allows the definition and use of pre-programmed acquisition sequences. For safety reasons, it is possible to bypass a sequence with manual control of the peristaltic pump or the detector.
Device characterisation
For all acquisitions, pulse height histograms with 256 bins were obtained. The histograms were energy-calibrated with a 137Cs source of 32.7 kBq, and the energy threshold was set to 110 keV for all acquisitions. An estimation of energy resolution was obtained by fitting a Gaussian on the photopeak (662 keV) of the energy spectrum.
Stability over time and catheter positioning reproducibility
Two series of tests were performed to evaluate the stability of the detector over time, both conducted with a 137Cs source (half-life of 30.17 years). The first one was used to verify that there is no drift in count rates over a period of 6 h with an integration time per sample of 5 s. The second test consisted in performing 19 acquisitions of 3 min at random times over a period of 3 weeks with a counting time per sample of 1 s. A 3D-printed template, shown in Fig. 6, was used to ensure reproducible positioning of the source relative to the detector. The template allows positioning of the source every 15 mm from the detector, with the closest position at 3 mm.
The first test over the 6 h acquisition was analysed by fitting the data to a linear function. A χ
2 test was used to verify that the data was Poisson-distributed, as expected.
In the second experiment, an analysis of variance (ANOVA) was performed to determine if the 19 acquisitions over 3 weeks belonged to a distribution with identical parameters (mean and variance).
Another experiment was conducted to verify that the catheter can be positioned in a reproducible manner each time. For a catheter filled with FDG, the catheter was removed and repositioned 20 times and counting was performed for 30 s with a counting time per sample of 1 s. Counting rates were decay-corrected, and an ANOVA was performed to assess reproducibility of catheter positioning. All statistical tests were performed with R (version 3.2.1).
Minimum detectable activity
The minimum detectable activity (MDA) is a crucial characteristic of the device; the counter must detect a small number of counts per second in the relatively high background of a PET scan room. This can be achieved through adequate shielding, coincidence counting or a combination of both. In this work, shielding only was used but the acquisition system of the device was designed with coincidence counting as an optional feature. Typically, the blood activity is lower than 500 kBq/ml at the maximum of the input function for fluorodeoxyglucose (FDG) [7].
The MDA, in units of kilobecquerel per millilitre, is defined at a 95% confidence interval by [27]:
$$ \text{MDA}= \frac{4.65\sqrt{N_{B}}+2.71}{fTs} $$
(1)
where T is the counting time per sample, N
B
is the number of background counts recorded during T, f is a factor of radiation yield per disintegration (f = 0.967 here) and s is the sensitivity of the detector as obtained by calibration. The sensitivity—the ratio of recorded count rate and activity concentration—was obtained with a 1.58-mm-inner-diameter catheter filled with 52.5 kBq/ml of FDG. The counting time per sample T used for MDA determination was 3 s, but it can be adjusted between 1 and 30 s to optimise the MDA as a function of background and activity level in the catheter. Figure 7 shows the prototype with FDG circulating in a catheter.
Background counts were measured in a realistic environment, i.e. in a PET scanner room at 1 m from a patient injected with 275 MBq of FDG 40 min prior to the experiment. The background count rate was averaged over a 3 min acquisition. The experiment was repeated with and without the tungsten shielding to estimate its efficacy.
Dispersion
To evaluate the activity dispersion in the tubing used, a step function study was conducted. A three-way valve was added at the end of the tubing (inner diameter of 1.58 mm) in a setup similar to the one used by Munk et al. [20]. Two vials were connected to the valve, one filled with water and the other with a mixture of water and FDG. Measurements were performed at two pump rates (2 and 4.5 ml min −1) and two tubing lengths (80 and 45 cm). The rising part of the step was modelled by an exponential function, f(t) = A(1− exp−t/τ) for each case [19].
PET study
The use of the device in real conditions is essential to verify that it meets clinical requirements and workflows. For this purpose, the device was tested in a clinical setting with prostate cancer patients undergoing dynamic 18F-fluoromethylcholine (FCH) PET studies. As shown in Fig. 8, the patient was in supine position with his arms positioned above his head. Two venous accesses were installed, one in each arm. FCH was injected in the left arm while blood was withdrawn from the right arm (18 gauge needle). Tubing of 76 cm with an inner diameter (ID) of 2.54 mm was connected between the patient and a stopcock. The stopcock interfaced a saline syringe and the tubing (ID of 1.58 mm) going to the P625 pump, the detector module and then the waste container. The blood withdrawal rate was set to 2 ml/min for the acquisition. The patient was injected with a standard activity of 4 MBq/kg for a total of 355 MBq. The injection was performed within 3 s. The pump was started approximately 1 min before the beginning of the acquisition, defined as the time where the FCH is injected while the PET acquisition and the gamma counting are started simultaneously. The dynamic PET scan with a field-of-view centred on the prostate lasted 600 s. To compensate for the transport delay of the blood in the tubing and to extend gamma counting time, 125 s were added to the gamma counting acquisition.
