Phantom‐based assessment of motion and needle targeting accuracy of robotic devices for magnetic resonance imaging‐guided needle biopsy

The current study proposes simple methods for assessing the performance of robotic devices intended for Magnetic Resonance Imaging (MRI)‐guided needle biopsy.

computed tomography are typically employed for localising the suspicious lesion and guiding the needle manipulator based on image feedback. 1,2 Robotically assisted biopsy has been introduced in multiple organs including the lung, 3 breast, 4 prostate, 5 liver, 6

etc.
Breast cancer is the most common form of cancer malignancy in women and the fifth cause of cancer-related death worldwide. 7,8 Diagnosis at an early stage provides patients with the best management and increased survival rates. Following clinical examination and imaging assessment, percutaneous biopsy is required for lesions suspicious for malignancy.
US is the modality of choice for image-guided biopsy of soft lesions that are visible on the US and is usually performed free-handed using core biopsy sampling. 9 Various robotic systems [10][11][12][13] for USbased needle guidance have been developed at the experimental level through the years to facilitate doctors. 14 Stereotactic mammography guidance is usually the gold standard for tumours that are not visible on US images, such as microcalcifications. Accordingly, there are several commercially available robotic mechanisms for needle alignment under the specific modality. 15,16 MRI is increasingly being used as the guidance modality for breast biopsy, especially for women with a high risk of breast cancer. 17 Since small lesions may not be detectable with the standard methods, 18 MRI-guided biopsy is needed to histologically determine the possible malignancy of the tumour. A wide variety of robotic systems that can be utilised in combination with an MRI scanner for automatic positioning of the needle have been developed so far, 4,[19][20][21] thus addressing the requirement to move the patient out of the scanner multiple times and pointless acquisition of a large tissue volume, as well as reducing the procedure duration.
These systems address challenges regarding safe operation of robotics in the MRI environment.
Such robotic-assisted approaches require high accuracy and precision in order to reach a target with minimal invasion and fulfil the clinical requirement. All methodologies for testing a robot's mechanical accuracy as well as the accuracy and repeatability of positioning a biopsy needle are based on comparing the intended with the actual motion step as determined using a distance-measuring method. 22,23 Before a process can be implemented in in vivo applications, its accuracy is typically assessed in free space, also known as the intrinsic system accuracy. Following demonstration of appropriate motion accuracy and repeatability through benchtop testing, the system should be employed in the real environment (e.g., an MRI system) to ensure that a high level of accuracy is maintained. 22 Several motion tracking methods were employed in the context of evaluating the targeting accuracy in benchtop experiments. 20,[24][25][26][27][28] Optical tracking devices have been frequently utilised in this regard, where the error of positioning was defined as the divergence of the actual from the intended location of the needle tip. 20,[24][25][26] As an example, a robot designed for breast biopsy was tested by navigating a rigid tool to reach target positions following both straight and curved pathways and tracking its motion using an optical tracker. 20 A similar approach was followed in a study by Groenhuis et al.,4 in which the needle positioning accuracy of a robot proposed for breast biopsy was tested in free air utilising a board with multiple crosshairs, which served as the targets. The needle tip was instructed to puncture these targets, and the positioning error was calculated by measuring the distance between the centre of each target and the respective punched hole. 4 High-quality breast biopsy phantoms constitute a valuable tool in needle biopsy training and experimentation, as well as in testing the needle positioning accuracy of robotic-assisted biopsy devices.
Commercially available breast biopsy phantoms are composed of patented realistic and durable breast tissue with several types of masses of varying sizes interspersed randomly throughout them. [29][30][31][32] Membranes mimicking the skin are utilised to cover the tissue, providing realistic needle resistance to trainees. [29][30][31][32] Not only commercially available, but also 'laboratory' phantoms have been proposed over time for biopsy training purposes. In fact, many institutions preferred home-made alternatives for their studies, utilising food or animal products, since they are low cost and easy to make. Turkey or chicken breast, as well as gel-based tissue-mimicking phantoms are commonly used to simulate soft tissue. 33 The inserts could be of liquid or solid content depending on the type of biopsy procedure that will be followed. For freehand US-guided breast biopsy training, raw materials including olives with pimentos, grapes, capers, strawberries, peas and potatoes have been proposed in the literature as tumour mimics. 33 Household and raw materials were also utilised for mimicking lesions in Magnetic Resonance (MR)-compatible breast phantoms. In that case, the success of the biopsy procedure can be confirmed directly by visualisation of the tumour material in the obtained sample 34 or through MRI by the location of the needle or the created void in relation to the target on intra-operative or follow up MRI images. 35 For instance, Werner et al. 34 assessed the reliability of an MRI-guided biopsy procedure in agarose phantoms containing peas, where the biopsy success was defined as the visual inspection of the specimen. Similarly, the accuracy of an automated system designed for lesion localization while operating at the MRI isocenter was demonstrated in a grapefruit phantom, in which the artificial lesions were vitamin E capsules, 36 by checking whether the capsule material was contained in the specimen chamber. Schneider et al. 35 also assessed an MRI-guided breast lesion localization system in a grapefruit-based phantom, in which breast tissue was mimicked by a grapefruit and the tumour by an embedded long wooden dowel. The success of the procedure was defined by whether pieces of the wooden dowel were contained within the sample, as well as from the void showed up on follow-up MRI. 35 Polyvinyl-alcohol cryogel (PVA-C) was also utilised to imitate the breast tissue in MRI-guided biopsy procedures. 4,37,38 The benefits of PVA-C include its robust mechanical properties, 39 as well as its ability to mimic the acoustic and MR relaxation properties of human tissue, and thus, to offer realistic visualisation on both MRI and US. 37 An indicative example is a study by Groenhuis et al.,4  3D-printed dedicated moulds. 4 Lesions with a size varying from 5 to 20 mm were simulated by fish oil capsules or pieces of stiff PVC added in the phantoms during the cool-off procedure. Various phantom sites were targeted by the needle based on pre-operative MR images and the error was estimated by the offset between the targeted site and reconstructed needle position. Following insertion, the needle was clearly visualised in MRI scans as a hole in the phantom. 4 Gel phantoms based on agar and gelatin have been proposed as a valuable cost-effective tool in the process of breast biopsy training under ultrasound needle guidance too. 40 In this effort, a recent study presents the development of an agar-based phantom mimicking breast, whose Young's modulus and acoustic properties were tuned to match those of fat, glandular and tumour tissues, thus providing realistic haptic feedback and US visibility during needle insertion. 41 A double-layered gelatin-based phantom with simulated tumours have been recently proposed for similar purpose in another study. 33 Interestingly, a glove finger filled with coloured solution served as a cyst, whereas the benign and malignant tumours were simulated by plant-based raw materials. 33 Agar phantoms could also constitute a valuable testing tool for MRI-guided robotic-assisted biopsy systems and protocols, mainly owing to its tissue-like MRI properties. [42][43][44][45][46][47][48] In fact, agar gels are predominantly chosen for MRI phantom studies because they can generate tissue-like MRI signal 42 and replicate the MR relaxation properties of different types of soft tissue upon addition of proper concentration of supplementary ingredients. 49 Furthermore, MRI phantoms are typically employed in the process of MRI safety testing of such devices, 50 which is required before the accuracy and precision of needle targeting can be actually tested in the real environment. Of note, Gadopentetate dimeglumine (Gd-DTPA); a commonly used contrast agent for MR imaging, 51 can be added in gel-based lesion mimics to improve the MRI contrast with respect to the surrounding breast tissue-mimicking material. 52 Either commercially available or home-made phantoms utilising household raw and laboratory materials are of great use in testing the performance of robotic-assisted breast biopsy devices and procedures. Herein, we propose some simple and cost-effective methods for testing MRI-guided breast biopsy robotic devices in terms of the accuracy of mechanical motion and needle targeting in both laboratory and MRI settings. In-house made dedicated phantoms containing agar as the gelling agent and biopsy targets served as the main tool in the evaluation process. An MRI compatible 2degrees of freedom (DOF) positioning device developed previously by our group was utilised in all the experiments. 53 Signal to noise ratio (SNR) experiments were initially carried out to confirm the safe and proper operation of the system in a 3T MRI scanner.
Planning for needle navigation relative to the target was performed on MRI images utilising the tools of a custom-made biopsy software. Laboratory evaluation of the needle targeting accuracy was based on a repeatability phantom test and a laser based-method.
The accuracy of needle targeting was then assessed by MRI phantom studies.

| Robotic positioning device
A positioning system previously developed by our group was employed for the purpose of the current study. 53 The system was manufactured on a 3D printing machine (FDM400, Stratasys) using Acrylonitrile Butadiene Styrene thermoplastic. The positioning mechanism features two DOF with a 6 cm motion range each, which are driven by piezoelectric motors (USR30-S3, Shinsei Kogyo Corp.) while motion feedback is provided by optical encoders (EM1-0-500-I, US Digital Corporation). The mechanical components were enclosed within a compact enclosure, resulting in the development of the device shown in Figure 1. As shown in the computer-aided design drawing of Figure 1A, a needle holder served as the end effector of the device extending from the mechanism through an arm.
The robotic device was designed based on two main requirements: (a) MRI compatibility and (b) highly accurate motion and needle positioning. In terms of MRI compatibility, the robot's materials and mechatronic components were carefully selected to ensure no interference with the scanner and safe operation of the robot in the MRI bore. A dedicated design is also required for the robot to fit the scanner while leaving sufficient space for comfortable patient placement, given the restricted space of the bore. The compact dimensions of the device of 40 cm in length, 15 cm in width and 7 cm in height enable its placement inside the MRI scanner between the subject and the bore for lateral approach of the needle, as shown in Figure 1A,B.
Regarding the second requirement, the targeting accuracy should be sufficiently high for targeting early stage cancer with the required precision and accuracy. A detailed description of the system design and manufacturing can be found in the study by Drakos et al. 53

| Needle navigation software
Navigation software integrating commands for MRI interfacing and robotic positioning control was interfaced with the robot for the purpose of the study. The software includes tools for navigation planning on pre-operative MR images and automatic positioning of the biopsy needle relative to the target area in the X-Y plane (robot coordinates). Figure 2A is a screenshot of the software interface for navigation planning.
An image-based registration approach involving the use of a water-filled syringe was followed to pre-operatively register the ro-  Communication with the mechatronic parts of the positioning mechanism is achieved through the electronic driving system located outside the MRI room through shielded cables. The extracted motion commands are thus sent to the electronic driving system for execution so that the syringe moves from its initial position to align with the target location (in a sagittal plane). Therefore, the entire procedure is conducted remotely in the control room, thus streamlining the biopsy process and increasing the overall efficiency. Figure 3 illustrates a hardware wiring diagram indicating the connection and flow signal among components.

| Agar-based biopsy phantoms
Phantoms simulating breast and tumour tissue were developed to be appearance, (c) excellent delineation of tumour mimics, and (d) proper stiffness both in terms of achieving reusability and providing tissue-like haptic feedback to the user during needle insertion. Agar was selected as the main ingredient of the phantoms considering the low cost and ease preparation of agar gels as well as their ability to emulate critical properties of various body tissues, including MRI properties. 42 A single-target breast-shaped phantom was prepared by embedding a cherry tomato in pure agar (Merck KGaA) gel of 4% weight per volume (w/v) concentration. The realistic shape of the phantom was achieved by pouring the agar mixture into a dedicated 3D-printed mold having a breast-shaped cavity. Figure 4A,B show photos of the developed phantom. Figure 4C shows a T2-Weighted A multiple-tumour phantom having a square shape and six embedded cylindrical targets of 10 mm in diameter was created by molding in a specially designed mold. Two agar mixtures with notable difference in the MRI relaxation times were prepared to mimic tumour and breast tissue. The concentration of materials for each phantom compartment was selected taking into consideration the T1 and T2 relaxation times of various agar/silica mixtures as measured in a previous study of our group, 49 in order to achieve good MRI contrast. Specifically, an agar gel was prepared using 6% w/v agar and 2% w/v silicon dioxide powder (Sigma-Aldrich) to mimic breast tissue, whereas another mixture of 6% w/v agar and a higher silica concentration of 6% w/v was prepared to mimic tumour tissue. Figure 5A shows a cross section view of the phantom, whereas Figure 5B

| Motion accuracy assessment
The accuracy of motion in the X-and Y-axes was evaluated using a digital calliper with a measuring resolution of 0.1 mm, which was

| MRI compatibility assessment
The next step in the evaluation process was to access the MRI compatibility of the robotic system. The experiment was carried In each case, the phantom was positioned at the isocenter of the magnet (0,0) and the signal was calculated as the average intensity in a circular ROI of 5 mm diameter, which was consistently placed at the same location in the phantom/air at a specific distance away from the coil, thus avoiding inhomogeneities due to inconsistent coil placement.
For each activation state, MR axial images were acquired with a

Repeatability phantom test
The purpose of this experimental part was to assess both the accuracy and repeatability of needle targeting in a developed phantom containing a cherry tomato, which served as the target. Needle navigation was performed 10 times to check the repeatability. Before each repetition, the needle holder was commanded to return to its home position.

Laser based method
The current experiment was carried out in another laboratory setting involving the use of a plastic grid instead of a phantom.

| Motion accuracy assessment
The measured range of actual displacement for each commanded distance (1, 5, 10, and 40 mm) in both the X-and Y-axes is listed in Table 1. Figure 9 shows the mean-measured distance (n = 20) versus the intended distance for the X-axis right and left directions, while Figure 10 shows the corresponding graph for the Y-axis upward and downward directions. The results indicate a maximum error of 0.1 mm for linear motion steps of up to 10 mm, whereas a two-times larger maximum error was observed for the motion step of 40 mm. driving system, as shown in Figures 11 and 12. Specifically, Figure 11 shows the axial 2D FLASH and T2-W SE images acquired at each tested state. A bar chart of the relevant SNR measurements for both tested sequences is shown in Figure 12.

| Benchtop evaluation
Regarding the repeatability test, the needle pierced the targeted tumour successfully in all 10 repetitions. Furthermore, there was a very good agreement in the location of the needle tip among repe- titions. An indicative photo showing the punctured tomato is shown in Figure 13. Figure 14 shows indicative results of the laser-based experiment, where a randomly selected cell located at the lower F I G U R E 1 0 Mean value of measured distance plotted against the intended distance for the Y-axis upward and downward directions. Error bars represent the standard deviation of the mean.
F I G U R E 1 1 Axial 2D FLASH and T2 W SE images of the breast-shaped agar-based phantom acquired at different activation states of the electronic driving system.

F I G U R E 1 2
Bar chart of the signal to noise ratio measured at different activation states of the electronic driving system using FLASH and T2-W SE sequences.

| MRI evaluation
Typical results of the needle targeting accuracy in the MRI setting are presented in Figures 15 and 16. After localization and placement of the syringe relative to the cherry tomato, T2-W sagittal SE images at the level of the syringe and the phantom were fused into the image of Figure 15A, which confirmed successful placement. The syringe was aligned with all the tumour mimics successfully, with a maximum offset between the commanded and actual syringe tip location of 1 mm. Successful puncture of the tumour was followed in all cases by manually advancing the needle. Figure 15B shows a T2-W SE image acquired after needle insertion, where the needle tip is visualised as a spot of reduced intensity within the tomato.  Figure 16A without any noticeable susceptibility artefacts. Figure 16B is a photo of the punctured phantom.

| DISCUSSION
In the current study, we aimed to share our experience in assessing It is now generally accepted that agar phantoms can provide a tissue-like signal in MRI. 42 In this study, a cherry tomato was selected as the first target since it has high water content appearing brighter than the surrounding tissue, whereas the MRI appearance of the agar-based tumour mimics was differentiated from the surrounding by adding a higher silica concentration. Notably, in case a fine-needle aspiration or a core biopsy needle will be utilised instead, the breast shaped phantom embedding a cherry tomato will allow for direct confirmation of the success of the sampling procedure through the visualisation of tomato pieces within the sample.
According to the tumor, nodes, and metastases classification for clinical breast cancer staging, invasive breast tumours are categorised into stages 1 through 4. 55 In terms of size, the first stage (T1) concerns tumours that are 20 mm or smaller in size at their widest area. 55 In this study, the authors selected a cherry tomato of almost 20 mm in diameter, whereas the agar-based tumour simulators had a smaller diameter of 10 mm. Therefore, the tumour models used can be considered representative of T1 stage tumours at least in terms of size. Although simplified tumour models of spherical/ellipsoid shape were employed in the study, one could easily create tumour models of a more complex shape using 3D printed dedicated molds.  Although simple and straightforward, this method has the limitation that it can only provide results with a mm resolution.
Needle navigation relative to the developed biopsy phantoms was also performed successfully in the MRI setting without any reported mechanical malfunction. The method involved the use of a water-filled syringe which was navigated to align with the location of a tumour mimic (in X-Y robot coordinates), which was then punctured by manually pushing the needle forward. The needle tip was clearly visualised within the target in follow-up T1-W and T2-W SE images with minor susceptibility artefacts. Remarkably, a small offset between the position of the syringe and needle was observed in the MRI scans and was attributed to the error introduced by the user.
These results further demonstrated that the system maintains accurate positioning when operated in the real environment and could be utilised for image-guided robotic-assisted biopsy procedures.
The robotic system employed in the study was designed in a

| CONCLUSIONS
The current study proposes ergonomic and cost-effective methods for assessing the performance of robotic devices intended for MRIguided needle biopsy in both laboratory and real environments.