This study was conducted in accordance to the guidelines of the local institutional review boards.
A three-step protocol study was performed, as follows:
-
(i)
to investigate the targeting accuracy in a rigid setting without organ motion, experiments were performed using a custom-made phantom of the trunk;
-
(ii)
to investigate the navigation system in vivo (considering respiration effect), a porcine trial was conducted under two different respiratory conditions, with and without breathing control;
-
(iii)
finally, to investigate the system accuracy in a quasi-clinical scenario, a cadaver with liver metastases was used.
We used a 5 mm accuracy threshold, defined as the distance between the geometric centre of the target and the needle tip, based upon key clinical considerations. In fact, most thermal ablations are performed on tumours with a diameter in the range of 1–3 cm [6, 7]. Thus, a guidance system must provide a targeting accuracy in the range of 5 mm from the geometric centre of the target in order to allow to achieve complete tumour necrosis in a single treatment session and to avoid destroying too much healthy tissue.
In all the three experimental settings, we computed accuracy and we verified that this limit was fulfilled.
Endosight system
To obtain augmented reality, a customised needle handle manufactured by using geometrically configured, fused deposition modelling technology attached to a 17 cm needle and markers glued on top of the handle were used. In addition, radiopaque tags were applied on patient’s skin to serve as fiducial markers. The augmented reality was then displayed by a tablet (Microsoft Surface Pro 4 with a 12-megapixel camera, Microsoft, Redmond, WA, USA) attached to a stable tripod platform (Fig. 1).
The workflow to generate augmented reality during the procedure is summarised in Fig. 2. Before intervention, radiopaque skin markers should be positioned on the patient’s body. Next, a CT scan is acquired. The CT images are then processed: automatic image segmentation algorithms are used to automatically obtain outline of target volumes. The fiduciary markers on the patient skin are segmented as well, with their position coordinates extracted using a principal component algorithm. During the intervention, the pre-treatment information, together with the real-time position of needle handle and patient markers, permits computation (the software used to create the augmented reality is: Unity 2017, f 1.1) and display of the augmented reality superimposed upon the visualised background of the interventional procedure. In fact, the proprietary software recognises a geometric configuration of fiducial markers on both the ablation guide handle and the patient. When the distance between the needle tip and the geometric centre of the target is 0, the needle is properly positioned at the centre of the intended target.
Anthropomorphic trunk model protocol
A semi-transparent silicone anthropomorphic phantom (50 × 34 × 27 cm) was constructed pouring a silicone material into a trunk gypsum mould (Fumagalli & Dossi, Milan, Italy). Multiple targets were positioned inside the phantom. These consisted of five polyvinyl chloride bars (two of 16 cm in length and 3 cm in diameter and four of 45, 30 or 20 cm in length and 2 cm in diameter). When the silicone hardened, the phantom was extracted by the mould. Thirty 3.5 × 3.5 cm radiopaque squares were placed on the phantom surface to serve as fiduciary markers, as shown in Fig. 3.
The phantom underwent CT using a 64-slice unit (LightSpeed, General Electric Healthcare Milwakee, WI, USA), with the following technical parameters: collimation = 3–5 mm; reconstruction interval = 2 mm; matrix = 512 × 512; in-plane pixel size = 0.48–0.86 mm; 120 kVp; 200 mAs; gantry rotation time = 0.8 s; and pitch = 1.75. Targets and markers were segmented and reconstructed in 3D with marker coordinates automatically derived.
To compute accuracy, we manually measured the distance between any pair of targets (geometric centre of bar) both on the actual phantom and on the augmented reality images obtained at three different tablet camera-phantom distances (30 mm, 40 mm, and 50 mm).
Porcine model protocol
Institutional animal care and use committee approval was obtained for the use of a swine model of this study. A female Yorkshire swine (aged 6 months old, weighing 93 kg; Mizra, Lahav, Israel) was studied with the supervision of the division of animal facility authority at Hadassah Hebrew University (Jerusalem, Israel). The animal received appropriate care from properly trained professional staff in compliance with both the Principles of Laboratory Animal Care and the Guide for the Care and Use of Laboratory Animals, approved by the Animal Care and Use Committees of The Hebrew University and in accordance with National Institutes of Health guidelines. The swine was initially anaesthetised with ketamine injection United States Pharmacopeia (Ketaset CIII, Fort Dodge Animal Health, IA, USA) 15 mg/kg, xylazine 2 mg/kg (Kepro, Deventer, The Netherlands), propofol 2% (one injection of 5 mL). Thereafter, the animal underwent endotracheal intubation followed by inhaled anaesthetic isofluorane (5% induction, 1.5–2.5% maintenance). No paralytics were used during the procedure.
The swine was placed on the CT table in prone position for kidney targeting and in the decubitus position for liver targeting. Using a coaxial 18-G needle technique (BrachyStar needles, Bard, Covington, GA, USA) under direct CT guidance, small (2 × 1 mm) metallic markers were embedded in three different anatomic locations (one in the kidney, two in the liver), as previously reported [8]. Twelve radiopaque markers were placed on the pig’s skin (Figs. 4a and 5a).
Once all targets were placed, a CT scan was performed with respiration suspended at maximum expiration. Then, the volume targets (skin, bone, targets, markers) were reconstructed by using reconstruction proprietary algorithms. Next, the needle was inserted using Endosight augmented reality guidance into the pig reaching each target centre indicated as a distance equal to 0 by the augmented reality system. Once completed, an expiratory breath-hold CT scan with the inserted needle was performed to verify the correspondence between what was shown by the augmented reality system and the result of the CT scan.
All CT scans were performed using a dual-layer 64-detector CT prototype, model iCT SDCT (Philips Healthcare, Cleveland, OH, USA). The scanning parameters were: 120 kVp; 200 mAs; gantry rotation time = 0.33 s; collimation = 40 mm (64 × 0.625 mm); and pitch = 0.984. Scans were reconstructed as contiguous slices of 1-mm thickness, using a standard soft-tissue convolution kernel.
The accuracy was simply measured as the distance between the geometric centre of the target and the needle tip on CT images. Two different conditions were tested:
-
(i)
with breathing control – the needle was inserted in the pig when it was in the maximum expiratory phase that corresponded to the same respiratory phase as that registered during the initial CT scan;
-
(ii)
without breathing control – with the pig breathing freely and the respiratory phase was not known.
Cadaver model protocol
This study was performed at the Simulation Center of Humanitas University (Pieve Emanuele – Milan, Italy) on a 73-year-old female torso cadaver with a history of liver metastases donated to science and obtained from medcure.org (Medcure, Orlando, FL, USA). To investigate location and number of liver metastases, the cadaver underwent ultrasound (US) scan (My Lab Gamma, Esaote, Genoa, Italy).
Based upon the US study, two liver metastases were selected as targets, one in segment IV (1.8 cm in size) and one in segment VI (3 cm). The cadaver was placed on its back on the CT gantry and a ventilator (Servo 900 C, Siemens, Erlangen, Germany) was attached to its trachea to induce simulated respirations. Twelve fiduciary surface markers were placed on the cadaver’s skin (Figs. 6a and 7a).
With the ventilator stopped midway in the respiratory cycle between inspiration and expiration, a CT scan of the cadaver was performed using a 16-slice unit (Emotion, Siemens Healthcare, Erlangen, Germany) with the following technical parameters: collimation = 3–5 mm; reconstruction interval = 2 mm; matrix = 512 × 512; in-plane pixel size = 0.48–0.86 mm; 130 kVp; 220 mAs; gantry rotation time = 0.5 s; and pitch = 0.784).
The volume of the targets (skin, bone, targets, markers) were then reconstructed using reconstruction proprietary algorithms. Subsequently, the needle was inserted in the cadaver using augmented reality guidance, until reaching the targets. The needle insertion was performed with the ventilator stopped in mid respiration to match the same respiratory phase of the initial CT scan until the target was reached on the augmented reality screen. Finally, CT scan of the cadaver with the inserted needle was performed. Accuracy of the final needle position was measured on CT images as for the pig model.
Statistical analysis
Data are presented as mean ± standard deviation (SD) or as mean and range.