|Year : 2020 | Volume
| Issue : 1 | Page : 24-31
Magnetic resonance imaging-guided focused ultrasound robotic system with the subject placed in the prone position
Christakis Damianou1, Marinos Giannakou2, George Menikou3, Leonidas Ioannou4
1 Department of Electrical Engineering, Cyprus University of Technology, Limassol, Cyprus
2 Department of R and D, Medsonic Ltd., Limassol, Cyprus
3 Medical Physics Division, Nicosia General Hospital, Nicosia, Cyprus
4 Department of Radiology, Ygia Polyclinic, Limassol, Cyprus
|Date of Submission||20-Jan-2020|
|Date of Decision||24-Mar-2020|
|Date of Acceptance||13-Apr-2020|
|Date of Web Publication||26-Aug-2020|
Department of Electrical Engineering, Cyprus University of Technology, 30 Archbishop Kyprianou Street, 3036 Limassol
Source of Support: None, Conflict of Interest: None
Background: In this article, a medical robotic system that performs magnetic resonance imaging-guided focused ultrasound surgery ablation is presented. The main innovation of this robotic system is that all the actuators are placed outside the water container. The transducer is immersed in water through an arm which is attached to the angular stage. Materials and Methods: The system includes three linear and one angular stage. The device uses piezoelectric motors for each motion stage. The accuracy was achieved with optical encoders. A focused transducer operated at 1 MHz with a radius of curvature of 10 cm and a diameter of 4 cm was used. A polyacrylamide gel was used to assess the ultrasound protocol. Results: The system was tested in the magnetic resonance imaging (MRI) environment and was proved to be a magnetic resonance compatible. The accuracy of the system was tested, and it was found that spatial steps of 0.2 mm can be safely and reliably achieved. With this robotic system, it is possible to access many organs that ultrasound penetrates with the patient placed in a prone position. Conclusion: The proposed robotic system can be modified so that it can be used for other applications. One example of an alternative application is MRI-guided biopsy. Another application is to replace the transducer arm with a radio frequency (RF) device to perform MRI guided RF ablation. Finally, the maneuverability of the robotic system can be enhanced further by attaching another angular stage to the system.
Keywords: Medical, magnetic resonance imaging, robotic, ultrasound
|How to cite this article:|
Damianou C, Giannakou M, Menikou G, Ioannou L. Magnetic resonance imaging-guided focused ultrasound robotic system with the subject placed in the prone position. Digit Med 2020;6:24-31
|How to cite this URL:|
Damianou C, Giannakou M, Menikou G, Ioannou L. Magnetic resonance imaging-guided focused ultrasound robotic system with the subject placed in the prone position. Digit Med [serial online] 2020 [cited 2021 Dec 8];6:24-31. Available from: http://www.digitmedicine.com/text.asp?2020/6/1/24/293503
| Introduction|| |
The first focused ultrasound surgery (FUS) system was built in 1940, and the general technology existed in an experimental setting for over 50 years. FUS can be focused on a targeted point to cause a rise in temperature between 60°C and 80°C. This can result in thermal tissue coagulation necrosis. Each sonication heats only a small focal target, so multiple sonications, must be used to ablate an entire target area. Due to the propagation mechanism of ultrasound, this technique should be performed with caution near nerves, bone, and rectum.
This technology has been employed in the nineties for prostate interventions. The FUS devices available for prostate cancer therapy are Ablatherm (EDAP TMS SA, Vaulx-en-Velin, France) and Sonablate  (Focus Surgery Inc., Indianapolis, IN, USA, now called SONACARE).
Recently, magnetic resonance FUS (MRgFUS) has been introduced due to a better ability to plan and monitor treatments in real-time. This technique is approved by the Food and Drug Administration for fibroid ablation. The InSightec systems have been deployed very fast in the last decade in other applications such as the treatment of prostate cancer, breast cancer, liver, and for pain palliation of bone metastases. Recently, Insightech developed a transcranial MRgFUS system, which eliminates noninvasively essential tremor. This application is becoming a primary choice of treatment and establishes FUS as a safe, effective, and noninvasive surgery. Therefore, the number of research institutions involved in the area of MRgFUS is growing rapidly. There is an unmet need for research institutions for affordable and functional preclinical robotic systems to explore new applications in MRgFUS.
There is an evolution of preclinical systems in the last decade, which is somehow slow. Chopra et al. 2009 developed an MR imaging (MRI) three-axis positioning system that delivers focused ultrasound in small animals for high-throughput preclinical drug delivery studies.
Another company that developed an MRgFUS system for small animal experiments is the French company Image Guided Therapy (Pessac, France). Image-guided therapy uses phased arrays to move the focus, and therefore, this technology is more expensive than the other existing technologies.
Another positioning device is the fully MR-compatible robotic assistance system InnoMotion (InnoMedic GmbH, Herxheim, Germany), which was originally designed for MR-guided interventions with needles. This system has five pneumatically driven degrees of freedom (DOF) and can be moved over a wide range within the bore of the magnet. The robotic system was combined with a fixed-focus ultrasound transducer. The size of this system is large; thus, the price of this system is high. The Canadian company FUS instruments  developed a 2-axis positioning device, with all moving parts immersed in water. This is a quite successful company regarding the commercialization of MRgFUS preclinical systems.
The proposed device can be applied for veterinary medicine, which always lagged behind human medicine. The proposed device offers benefits of performing clinical trials (only IRB is needed) in companion animals. Our dogs and cats are exposed to the same environmental stimuli that we are, and develop many of the same diseases in a far more natural way than laboratory animals. Veterinary trials make new innovative therapies available for family pets, while simultaneously collecting data that can be used to advance human medicine.
Focused ultrasound offers several advantages over traditional treatments such as surgery and radiation. It is noninvasive, which reduces the risk of infection and eliminates the need for stitches and the Elizabethan collar. Focused ultrasound can be used to ablate tissue or enhance the local delivery of therapeutic drugs. As there is no ionizing radiation involved, treatments can be repeated if needed. Focused ultrasound has many potential applications in veterinary medicine, including but not limited to tumor destruction, drug delivery (chemotherapy and immunotherapy), pain relief for arthritis, and hip dysplasia.
An MRI compatible positioning device is proposed for the treatment of pet diseases. The device features 4 DOF that is used to maneuver a single element transducer; hence, it can be produced at a lower cost compared to the devices currently available. The robot includes three linear stages and one angular, and this design allows the delivery of the ultrasound through multiple angles; therefore, it can access multiple organs. Furthermore, it is suitable for any commercial 1,5T and 3T MRI scanner. A user-friendly software was developed that communicates with the MRI scanner, control the positioning device and the ultrasound system. In addition, the robot has small dimensions, and the overall system is very compact.
| Materials and Methods|| |
An MRI compatible positioning device with 4 DOF was developed for the treatment of pet diseases. The device is used to maneuver a single element transducer. The robot includes three linear stages and one angular. The device is suitable for any commercial 1,5T and 3T MRI scanner. An electronic driving system with an ultrasound biography (USB) interface was used to control the device. A user-friendly software was developed that communicates with the MRI scanner, control the positioning device and the ultrasound system.
The functionality of the ultrasound system was evaluated using polyacrylamide gel and agar-based phantoms. The phantoms were used to assess the ability of the ultrasound to induce heat in the tissue. MR compatibility tests were performed by measuring the SNR for various conditions. The ability to monitor the treatment progress was demonstrated using MR thermometry maps to detect the heating at the focus. Additional tests showed the motion accuracy and repeatability of the device.
Positioning device description.
The positioning device was designed using the Microstation (V8, Bentley Systems, Inc.) CAD software. The parts were manufactured using a 3D printer (FDM400, Stratasys, 7665 Commerce Way, Eden Prairie, Minnesota, 55344, USA). The device components were made of ABS material, which offers good mechanical properties; thus, the positioning device is robust and reliable.
The positioning device has 4 DOF, 3 linear axis, and 1 angular axis. Each axis is driven by a piezoelectric ultrasonic motor (USR60-3N, Shinsei Kogyo Corp., Tokyo, Japan). The dimensions of the device are 64 cm in length, 25 cm in width and approximately 29 cm (when Z-axis is fully extended) in height. Due to the small dimensions of the device, it can be used with any commercially available MRI scanner. The device has a motion range of 5 cm on the Z-axis, 7 cm for the X-axis, 7 cm for the Y-axis (relative to the MRI), and 180° (90° on each direction) for the Θ-axis. The weight of the positioning device is approximately 4 kg.
[Figure 1] shows the CAD representation of the robotic system. [Figure 1]a and [Figure 1]b shows the 3D drawings of the positioning device. [Figure 1]c shows the Z-stage (MRI axis) of the positioning device, which features a jackscrew mechanism for the conversion of the angular motion to linear motion. On the rear of the base, the piezoelectric motor is visible, which is coupled to the jackscrew mechanism.
|Figure 1: Three dimensional drawings of the positioning device (a and b). (c) the Z-stage (magnetic resonance imaging axis) of the positioning device, (d) X-stage, (e) Y-stage, (f). Θ-stage|
Click here to view
[Figure 1]d shows the X-stage drawings. The X-axis moves along the Z-plate guides. The ultrasonic motor was attached to the Z-plate to transfer the motion to the X-stage. On the rear of the X-plate an optical encoder was used to ensure the accurate motion of the X-stage. Two structures were added to strengthen the X-plate since the middle section was smaller to increase the motion range.
[Figure 1]e shows the Y-stage, which followed the same design principles with the Z-stage and the X-stage. The motor was fixed to the top of the Y-frame. On the front left side of the Y-frame an optical encoder was placed. On the front of the Y-plate a coupling was made to attach the Θ-stage [Figure 1]f. The theta stage mechanism differs from the linear stages. The motion ιs transferred through a shaft that includes a screw on the lower end. The screw is coupled to a gear that is attached to the lower arm. Using this technique, allowed the placement of the motor on the top of the Θ-stage, thus placing the ultrasonic motor far from the water container. An MR compatible ultrasonic transducer (Sonic Concepts, USA) was attached to the device using a coupling mechanism. The water was used to create an acoustical coupling between the ultrasound and the target. [Figure 2]a shows the manufactured positioning device and [Figure 2]b shows the positioning device as placed inside the MRI scanner.
|Figure 2: (a) Manufactured positioning device and (b) positioning device as placed inside the magnetic resonance imaging scanner|
Click here to view
To ensure motion accuracy, optical encoders were used on all the stages. Linear stages were equipped with optical encoders (EM1-0-500-I EM1, US Digital Corporation, Vancouver, WA 98684, USA) in conjunction with a polymer plastic strip (LIN-500-10-1, US Digital Corporation, Vancouver, WA 98684, USA) with the resolution of 500 lines pe inch. For the angular stage, an optical encoder was used (EM1-2-2500-I, US Digital Corporation, Vancouver, WA 98684, USA) in conjunction with a plastic disc (DISK-2-2500-093-IE, US Digital Corporation, Vancouver, WA 98684, USA) with a resolution of 2500 lines per 360°. The encoders were connected to a data acquisition (DAQ) board USB-6351 (NI, Austin, USA). This robot is classified as MR conditional, according to ASTM standards, since the encoder modules and the ultrasonic transducer require electricity to operate (F2503, F2052, F2213, F2182, and F2119).
The robotic system control software was written in C # (Visual Studio 2010 Express, Microsoft Corporation, USA). It offers a user-friendly interface and enables the user to control the robotic device motion, to activate the ultrasound amplifier and allows communication with the MRI scanner. The command history is stored so that the user can review the previous commands if necessary. The software is automatically updating the transducer position coordinates in relation to its initial location. The software includes the option to store patient information such as patient name, age, address health history, medical images, etc. In addition, an MR compatible camera (MRC Systems GmbH, Heidelberg, Germany) can be connected to acquire images. The software can connect to the MRI scanner using the Digital Imaging and Communication in Medicine (DICOM) protocol. The connection was established by entering the appropriate parameters for the server name, application entity title, IP address, and port. The software, when connected to the DICOM server, can download and store study identification (ID) or series ID using the patient identification (ID). The files are downloaded in a local server and stored in the computer.
Each ultrasonic motor is controlled by a driver that provides the appropriate sinewave signals to rotate it. A DC power supply (24 V, 6 A) is used to provide power to the drivers (D6060, Shinsei Kogyo Corp., Tokyo, Japan). The motion of the positioning device is controlled through a USB 6351 DAQ interface card (National Instruments, Austin, Texas, USA), which is hosted inside the electronic enclosure.
A polyacrylamide gel was used (ONDA Corporation, Sunnyvale, CA, USA) for the evaluation of the system. The polyacrylamide gel is transparent and when exposed to the high temperatures, it changes its color to white. The white lesions inside the gel appear as a solid hence, it allows visual inspection of the induced lesions and confirmation of the spatial accuracy.
Additionally, experiments were conducted with agar-based phantoms. The phantoms were made using 6% w/v agar, 4% w/v silica dioxide, 30% v/v evaporated milk. The evaporated milk, which is rich in fats and proteins, was used to adjust the attenuation while scattering ultrasound weakly. Scattering was compensated by adding to the mixture of appropriate amounts of fine crystalline silica dioxide powder. The percentage of each product was adjusted to mimic the acoustic attenuation (~ 1 dB/cm-MHz) of muscle tissue, as it was described by Menikou et al. The agar gel phantoms were used to evaluate the ultrasonic protocol. By using MR thermometry, it was possible to detect the focus of the transducer and to observe the temperature elevation at the focus.
Focused ultrasound surgery system
The FUS setup is composed of a spherically focused MR compatible ultrasonic transducer (Sonic Concepts, USA) operating at a frequency of 1MHz. The transducer has a radius of curvature of 10 cm and a diameter of 4 cm. The ultrasonic transducer is operated by a radio frequency (RF) generator (RFG 750 W, JJA, Seatlle, WA, USA), which is controlled by the software.
Experiments in gel phantoms
To test the functionality and repeatability of the positioning device, experiments were carried out with gel phantom. The phantom was placed on the front cover of the water container in the center of the opening. The ultrasonic transducer was attached to the arm of the positioning device using a coupling. The transducer and the bottom surface of the phantom were immersed in the water to enable the ultrasound to reach the phantom.
Magnetic resonance imaging
To assess the MR compatibility of the robotic system tests under different conditions were performed. The signal-to-noise ratio (SNR) was measured with the robot placed in a 1.5T MRI scanner (Signa, General Electric, Fairfield, CT, USA) with a GPFLEX coil (USA instruments, Cleveland, OH, USA) that enabled higher quality images of the phantom. The SNR was measured with the encoders activated and deactivated, with the presence and absence of the transducer and during activation of the transducer using a T1-weighted spoiled gradient (SPGR) sequence. The parameters used for the SPGR sequence was: Repetition time (TR) = 38.5 ms, Echo time (TE) = 20 ms, Field of view (FOV) = 21 cm, matrix =128 × 128, flip angle = 20°, Number of excitations (NEX) = 1.
Furthermore, the motion of the transducer was evaluated using Fast Recovery Fast Spin Echo (FRFSE) T2-weighted sequence with the following parameters: TR = 2200 ms, TE 61.2 ms, slice thickness =15 mm, matrix = 192 × 192, FOV = 17 cm, NEX = 1, and echo train length = 16. The high-resolution FRFSE images were acquired between each movement. By comparing the previous MR image transducer position to the current transducer position, the repeatability and motion accuracy of the positioning device was quantified.
Magnetic resonance thermometry
To observe the temperature elevation during FUS sonications, MR thermometry was used in order to monitor the rate in which the temperature built up. The temperature calculation was based on the phase shift proton resonance frequency shift phenomenon. By measuring the difference in phase shift between two MR images in a pixel by pixel basis, the temperature increasement was calculated.
To assess the thermometry at the focus, slices were acquired along the long axis and the short axis of the beam. This ensures the accurate detection of the peak temperature at the focus. The MR images for the thermometry were acquired using SPGR echo sequence. The images were processed by a custom-made software developed in MATLAB (MathWorks, Natick, United States). The software compares phase shifts developed during the acquisition of non-treated image (mask) and treated images. Temperature elevation was returned by the software as the maximum value in a prescribed region of interest (ROI) that was manually positioned. Temperature-color coded maps were produced by adjusting the color map (blue to red) for a range of minimum to maximum ROI temperature.
| Results|| |
[Figure 3] shows the results for the evaluation of motion for one of the linear axis (x). It shows the distance moved versus intended step in mm with typical steps used during navigation of the robotic system (1 to 5 mm). The test was repeated 20 times. The distance moved (red line) is very close to the expected distance (average difference was 0.06 mm at 1 mm step and 0.2 for the 10 mm step). The other two axes (Y and Z) revealed similar differences in accuracy.
|Figure 3: Evaluation of motion for one of the linear axis (x) showing distance moved versus intended step in mm with typical steps used during navigation of the robotic system (1–5 mm). The test was repeated 20 times|
Click here to view
[Figure 4] shows the results for the accuracy of the angular axis. The graph shows the measured angle in degrees versus intended angle in degrees (varied from 2° to 10° with n = 20). The angle moved (red line) is slightly bigger than the expected angle (average difference in degrees was 0.05° at 2° step and 0.1° for the 10°).
|Figure 4: Measured angle in degrees versus intended angle in degrees (varied from 2 to 10 degrees with n = 20) foe the angular stage|
Click here to view
[Figure 5] shows the SNR using SPGR measured in the agar-based phantom for different component activation. The SNR was maximized when the motor, encoder, and transducer were OFF. There was some drop in the SNR when the motor was energized and further drop in the SNR was observed when the motor was energized. The maximum decrease of SNR occurred when all components were activated (encoder, motor, and ultrasonic transducer) were energized.
|Figure 5: Signal to noise ratio using spoiled gradient measured in the agar-based phantom for different component activation|
Click here to view
[Figure 6] shows images using MR thermometry in a plane perpendicular to the ultrasonic propagation (coronal) at different time intervals (every 12 s) using acoustical power of 30 W for 60 s. [Figure 7] shows images using the MR thermometry map in a plane parallel to the ultrasonic propagation (sagittal) with the same ultrasonic exposure. This figure demonstrates the growth of the thermal beam of the transducer of the robotics-based system.
|Figure 6: Magnetic resonance thermometry in a plane perpendicular to the ultrasonic propagation (coronal) at different time intervals (every 12 s) using acoustical power of 30 W for 60 s|
Click here to view
|Figure 7: Magnetic resonance thermometry map in a plane parallel to the ultrasonic propagation (sagital) with the same ultrasonic exposure as in Figure 6|
Click here to view
| Discussion|| |
In this article, we presented a robotic system that performs MRgFUS ablation. The main innovation of this robotic system is that the motors, encoders, and moving parts are located outside the water container. The transducer is immersed in water via an arm which is attached to the angular stage. Therefore, the complexity of passing the transducer arm through water-proof mechanisms is avoided.
The system was tested in the MRI environment and was proved to be MR compatible. The accuracy of the system was tested, and it was found that spatial steps of 0.2 mm can be safely and reliably achieved.
With this robotic system, it is possible to access many organs that ultrasound penetrates. With the patient placed in prone position access is possible to the breast,, liver, kidney, and pancreas. Access will be possible to the brain provided that a phased array replaces the single element transducer. For the brain application, the patient may be positioned supine. Finally, for some cases of bone pain palliation, this system may be used.
The proposed robotic system can be easily modified so that it can be used for other applications. One example of an alternative application is MRI-guided biopsy. The transducer arm can be replaced with a needle biopsy. Another application is to replace the transducer arm with a RF device to perform MRI guided RF ablation. The possibilities for this system are many. Finally, the maneuverability of the robotic system can be enhanced further by attaching another angular stage to the system.
Financial support and sponsorship
Conflicts of interest
There are no conflicts of interest.
| References|| |
Lynn JG, Zwemer RL, Chick AJ, Miller AE. A new method for the generation and use of focused ultrasound in experimental biology. J Gen Phyciology 1942;26:179-93.
Jolesz FA. MRI-guided focused ultrasound surgery. Annu Rev Med 2009;60:417-430.
Blana A, Rogenhofer S, Ganzer R, Lunz JC, Schostak M, Wieland WF, et al
. Eight years' experience with high-intensity focused ultrasonography for treatment of localized prostate cancer. Urology 2008;72:1329-33.
Murat FJ, Poissonnier L, Pricaz E, Rouviere O, Mege F, Chapelon JY, et al
. Prognostic factors for salvage HIFU success after external beam radiation failure (EBRT) failure. Eur Urol Suppl 2008;7:119.
Warmuth M, Johansson T, Mad P. Systematic review of the efficacy and safety of high-intensity focussed ultrasound for the primary and salvage treatment of prostate cancer. Eur Urol 2010;58:803-15.
Yehezkeli O, Freundlich D, Magen N, Marantz C, Medan Y, Vitek S. INSIGHTEC-TXSONICSLTD, assignee. In: Mechanical Positioner for MRI Guided Ultrasound Therapy System WO0209812, Inventors Word Intellectual Property Organization; 2002.
Napoli A, Anzidei M, De Nunzio C, Cartocci G, Panebianco V, De Dominicis C, et al
. Real-time magnetic resonance-guided high-intensity focused ultrasound focal therapy for localised prostate cancer: Preliminary experience. Eur Urol 2013;63:395-8.
Khiat A, Gianfelice D, Amara M, Boulanger Y. Influence of post-treatment delay on the evaluation of the response to focused ultrasound surgery of breast cancer by dynamic contrast enhanced MRI. Br J Radiol 2006;79:308-14.
Kopelman D, Inbar Y, Hanannel A, Dank G, Freundlich D, Perel A, et al
., Magnetic resonance-guided focused ultrasound surgery (MRgFUS). Four ablation treatments of a single canine hepatocellular adenoma. HPB 2006;8:292-8.
Catane R, Beck A, Inbar Y, Rabin T, Shabshin N, Hengst S, et al
. MR-guided focused ultrasound surgery (MRgFUS) for the palliation of pain in patients with bone metastases-preliminary clinical experience. Ann Oncol 2007;18:163-7.
Lipsman N, Schwartz M, Huang Y, Lee L, Sanka T, Chapman M, et al
. MR-guided focused ultrasound thalamotomy for essential tremor: A proof-of-concept study. Lancet Neurol 2013; 12:462-68.
Chopra R, Curiel L, Staruch R, Morrison L, Hynynen K. An MRI-compatible system for focused ultrasound experiments in small animal models. Med Phys 2009;36:1867-74.
Krafft AJ, Jenne JW, Maier F, Stafford RJ, Huber PE, Semmler W, et al
. A long arm for ultrasound: A combined robotic focused ultrasound setup for magnetic resonance-guided focused ultrasound surgery. Med Phys 2010;37:2380-93.
Menikou G, Dadakova T, Pavlina M, Bock M, Damianou C. MRI compatible head phantom for ultrasound surgery. Ultrasonics 2015;57:144-52.
Menikou G, Damianou C. Acoustic and thermal characterization of agar based phantoms used for evaluating focused ultrasound exposures. J Ther Ultrasound 2017;5:14.
Melodelima D, Dupre A, Chen Y, Perol D, Rivoire M. Toroidal intra-operative high intensity focused ultrasound transducer for treating liver metastases under ultrasound imaging guidance: Clinical results of Phase II study. J Acoust Soc Am 2019;145:1862.
Brenin DR, Patrie J, Nguyen J, Rochman CM. Treatment of breast fibroadenoma with ultrasound-guided high-intensity focused ultrasound ablation: A feasibility study. J Br Imaging 2019;1:316-23.
Zhao J, Wang Q, Li X, Hu X, Shen H. Effective high intensity focused ultrasound treatment in recurrent aggressive breast fibromatosis: A case report. Onco Targets Ther 2019;12:5251-6.
Burks SR, Lorsung RM, Nagle ME, Tu TW, Frank JA. Focused ultrasound activates voltage-gated calcium channels through depolarizing TRPC1 sodium currents in kidney and skeletal muscle. Theranostics 2019;9:5517-31.
Razavi M, Zheng F, Telichko A, Ullah M, Dahl J, Thakor AS. Effect of pulsed focused ultrasound on the native pancreas. Ultrasound Med Biol 2020;46:630-8.
[Figure 1], [Figure 2], [Figure 3], [Figure 4], [Figure 5], [Figure 6], [Figure 7]
|This article has been cited by|
||Robotic system for magnetic resonance guided focused ultrasound ablation of abdominal cancer
| ||Anastasia Antoniou,Marinos Giannakou,Nikolas Evripidou,Georgios Evripidou,Kyriakos Spanoudes,Georgios Menikou,Christakis Damianou |
| ||The International Journal of Medical Robotics and Computer Assisted Surgery. 2021; |
|[Pubmed] | [DOI]|
||A Robotic Platform to Navigate MRI-guided Focused Ultrasound System
| ||Jing Dai,Zhuoliang He,Ge Fang,Xiaomei Wang,Yingqi Li,Chim-Lee Cheung,Liyuan Liang,Iulian Iordachita,Hing-Chiu Chang,Ka-Wai Kwok |
| ||IEEE Robotics and Automation Letters. 2021; 6(3): 5137 |
|[Pubmed] | [DOI]|