A. Sentinel lymph node mapping
Sentinel lymph node biopsy (SLNB) is a widely adopted technique for assessment of lymph node status in cancer patients. Several hours prior to surgery a subcutaneously injected radioisotope-labeled colloidal tracer, and sometimes a blue dye, is injected in the proximity of the tumor(s). The radiotracer particles drain via the lymphatic system and are trapped in nearby lymph nodes. During surgery, lymph nodes with the highest tracer accumulation (defined as sentinel nodes) are excised and histological node status analysis is performed immediately. If no malignant cells are found, no further nodes are removed, thereby minimizing postsurgical lymphadenectomy-related morbidity.
In current practice, initial visualization of general SLN location is provided by lymphoscintigraphy, in which a large, general-purpose gamma camera acquires planar scintigraphic images immediately following radiotracer drainage. During surgery a non-imaging handheld gamma probe, providing variable-pitch audio feedback to the surgeon, is used for SLN localization, excision, and ex-vivo activity measurement. SLNB is a routine component of surgical management and staging of patients with clinical stage IB and II melanoma, where the pathologic status of the SLN is the most important prognostic factor [1]. However, in a meta-analysis of more than 25,000 patients [2] using the combination of preoperative lymphoscintigraphy and intraoperative gamma probe, node localization false negative rates averaged 12.5%. In cases where the cause for a false negative is able to be identified, 44-50% have been ascribed to failure of radiologic or surgical identification of the sentinel lymph node [3].
Small field of view gamma cameras, typically mounted on a mobile gantry arm, provide a means to potentially reduce the number of false negatives in SLNB compared to the combination of pre-operative lymphoscintigraphy and intraoperative use of a non-imaging probe [4–9]. Such compact imaging devices can provide updatable, real-time 2-dimensional (2D) images of SLN location in the operating room and could be used immediately pre-surgery provide updated images to guide SLNB planning. Mobile gamma cameras have also been evaluated for use in radiomarker localization in procedures such as radioguided occult lesion localization and radioguided seed localization [10,11].
The declipse®SPECT system, also known as freehand SPECT (fhSPECT), was developed by SurgicEye (Munich, Germany) to provide pre-operative and intraoperative 3D imaging of radiotracers [12,13]. As originally developed and marketed, the system uses an overhead infrared imaging system to track the position and orientation in space of a non-imaging probe as it is raster scanned above the patient surface. Count rate data from the probe’s single detector are streamed to a fast reconstruction algorithm to generate a 3D (fhSPECT) image of the radiotracer distribution that is superimposed upon a live visible light video image of the patient. Declipse®SPECT (also referred to here as probe fhSPECT) has been widely used in SLN mapping in a variety of cancers including breast [14,15] and melanoma [16–18]. It has also been successfully employed in localization of pre-operatively placed radioactive tumor markers such as the 125I seeds used in breast cancer surgery [19]. Fitted with a laparoscopic extension for the gamma probe, declipse®SPECT has also been evaluated for guidance in excision of pulmonary nodules pre-operatively marked using 99mTc-macroaggregated albumin using CT guidance [20,21].
In this project, the non-imaging probe was replaced by a unique silicon photomultiplier (SiPM)-based handheld gamma camera developed in a collaboration among SurgicEye, the Thomas Jefferson National Accelerator Facility (Jefferson Lab, Newport News, VA), the University of Virginia, Dilon Technologies (Newport News, VA), and West Virginia University. The potential advantages of an imaging camera instead of a non-imaging gamma probe include a much larger detection surface area, the simultaneous use of multiple gamma-sensitive detector elements, higher spatial resolution, and more complete tomographic sampling. Here we describe the results of a feasibility study using the camera-based freehand SPECT system (referred to here as camera fhSPECT) for pre-operative SLN mapping among melanoma surgical patients.
B. Imaging System Description
The 7 cm diameter handheld gamma camera (Figure 1) incorporates a 6 mm thick pixelated thallium doped sodium iodide (NaI(TI)) scintillator, an array of 80 silicon photomultipliers (Hamamatsu Photonics model S10362-33-050P) with a pitch of 6 mm, and a custom-built tungsten-polymer composite parallel-hole collimator with hole size 0.6 mm and hole length 5.5 mm [22]. The central region of the NaI(Tl) crystal is pixelated to form a 25x25 array of 2.25 mm x 2.25 mm crystals with 2.5 mm pitch. The camera, designed and built at Jefferson Lab, is housed in a cylinder of CMW-1000 machinable tungsten with an interior layer of black Delrin® (polyoxymethalyne). This interior layer shields the SiPMs from ambient light and mechanically stabilizes the camera’s electronics. The combined mass of the camera and housing together is 1.4 kg, allowing it to be scanned by hand without mechanical support. The camera’s design and performance with a single-crystal lanthanum bromide scintillator has been described previously [22].
For 3D imaging, the rear surface of the gamma camera housing was fitted with an array of four reflective spheres (Figure 1) so that IR light from the freehand SPECT overhead tracking system can be used to determine camera location and orientation (Figure 2). A similar array containing three reflective spheres is placed on the surface of the subject being scanned in order to provide a fixed reference to the subject surface.
During image acquisition, the handheld camera is first swept in a non-imaging fashion over a broad region of the subject’s surface with the goal of identifying areas producing relatively high counting rates. Once high-count regions are identified, a more systematic raster scan is performed in imaging mode to image a cubic volume approximately 1,000 cm3 (10 cm on each side). Ideally, during the 3D scan the camera is moved to obtain projection images in two or preferably three mutually perpendicular planes. In many cases three precisely orthogonal camera orientations are impossible because of the node location relative to the surface topography of the subject. However, even in these cases useful 3D images of SLNs can be obtained with nearly isotropic spatial resolution by changing the camera orientation as far as possible during the scan. The total scan time is typically 90 seconds, at which point the Declipse software performs image reconstruction. Quantitative data on the depth and intensity of detected focal activity are displayed on a real time video image of the patient surface.
C. Laboratory characterization of imaging properties
a. Energy resolution
A low scatter source comprising a thin layer of 99mTc-pertechnitate solution in a thin plastic petri dish was used to acquire a pulse height histogram. The full width at half maximum of the 140 keV peak was measured to quantify the energy resolution of the handheld SiPM gamma camera. Averaged over seven trials the energy resolution was found to be 21.5 ± 1.7 % (mean ± 95% confidence interval) at 140 keV.
b. Spatial resolution
To characterize the 2D spatial resolution of the handheld camera, thin line sources were created with 99mTc-pertechnetate in capillary tubes (Kimble 71900-50 𝜇L, 1 mm inner diameter). The line sources were imaged at 10 mm source-to-collimator intervals beginning at the surface of the collimator until a distance of 100 mm was reached. At each location, the capillary was oriented at a small angle with respect to the crystal matrix so that the width of its image could be averaged over multiple offsets of the capillary from the crystal centers [23]. The phase-averaged FWHM of the width of capillary image in each 2D projection image was recorded, and the average of seven trials is plotted versus changing capillary-to-collimator separation in Figure 3.
The reconstructed 3D spatial resolution was measured by placing a drop of 99mTc-pertechnitate solution in the tip of an Eppendorf tube. The source was scanned in air using a source-to-camera separation of ~10 cm. The spatial resolution was measured in three perpendicular dimensions by extracting the central slice from the reconstructed volume. Averaged over three scans, the FWHM resolution was 11.9 +/- 2.5 mm, 13.3 +/- 0.3 mm, and 14.2 +/- 1.8 mm in the coronal, sagittal, and axial planes, respectively. These three planes were defined in terms of a supine surgical patient.
c. Sensitivity
In both 2D and 3D the sensitivity was defined as the ratio of total number of image counts per second and the source activity. Source size was chosen so that the entire source image fit within the field of view of the handheld SiPM gamma camera.
The 2D sensitivity of the gamma camera was experimentally determined according to the standards of the National Electrical Manufacturers Association (NEMA NU 1-2012 ‘Performance Measurements of Gamma Cameras’). A thin layer of 99mTc-pertechnetate in a 10 mm diameter petri dish was imaged with a source-to-collimator separation of 10 cm. The sensitivity was found to be 171 ± 6.2 cps/MBq.
The 3D sensitivity of the imaging system was evaluated by imaging Eppendorf tubes with activities in the range of 3.59 to 6060 mCi (0.132 to 224 MBq). A source to collimator separation of approximately 30 cm was used during scanning. 3D sensitivity is shown plotted versus source activity in Figure 4. The results show that the 3D sensitivity is uniform in this range of activity with a mean sensitivity of 203 ± 19.5 cps/MBq. The small difference between the 2D and 3D sensitivity is attributable to the use of a slightly more narrow energy window in 2D imaging compared to a wider energy window used for 3D scanning.
d. Node localization accuracy
After scanning is completed and 3D image reconstruction has been performed, the investigational handheld camera fhSPECT system interactively displays the separation between the gamma camera and regions of focal radioactivity, such as SLNs. If the camera is placed face down on the patient’s surface this provides a measurement of node depth. A series of tests were performed to determine the accuracy of the localization measurement reported by the imaging system. Such information is useful in surgical planning of optimal paths to nodes, and is not readily available with non-imaging gamma probes.
Eight nodes were simulated using small spheres filled with activities ranging from 18 µCi to 52 µCi (average of 26 µCi or 0.96 MBq). Following each node scan, the input surface of the camera was placed in contact with the node, then at a distance of 35 mm. The known camera-to-node separations were then compared to those reported by the imaging system, as shown in the bar plot of Figure 5. For the 16 trials (eight nodes and two post-scan camera positions each) the average error in the reported camera-to-node separation was 9.2 mm with a standard deviation of 2.7 mm. However, the separation reported by the system is actually that between the node and the input surface of the NaI(Tl) scintillator. That surface is separated from the camera’s outer surface by the thickness of the camera housing. Taking into account the combined 9 mm thickness of the collimator and camera housing, this corresponds to a true error of 0.2 ± 2.7 mm.
The accuracy of the system’s ability to measure the separation between nodes was evaluated using a phantom containing two point sources at known separations. Figure 6 shows a schematic of the phantom, which contains two acrylic posts of different heights, with a center-to-center spacing of 20 mm. A drop of 99mTc-pertechnetate solution was added to the small well in the center of each post’s top surface. The phantom was scanned and then the camera was placed at multiple locations ranging from ~5 cm to 20 cm from the sources. The reported source-to-camera separations were used to calculate the inter-source separation in the X, Y, and Z dimensions compared to the known values (20 mm, 0 mm, and 20 mm). The mean absolute error of the inter-source distance was 1.2 ± 0.34 mm (95% confidence interval).