Graphic Art

A Tele-Operated Mobile Ultrasound Scanner Using a Light-Weight Robot

A Tele-Operated Mobile Ultrasound Scanner Using a Light-Weight Robot
of 9
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
  50 IEEE TRANSACTIONS ON INFORMATION TECHNOLOGY IN BIOMEDICINE, VOL. 9, NO. 1, MARCH 2005 A Tele-Operated Mobile Ultrasound ScannerUsing a Light-Weight Robot Cécile Delgorge, Fabien Courrèges, Lama Al Bassit, Cyril Novales, Christophe Rosenberger, Natalie Smith-Guerin,Concepció Brù, Rosa Gilabert, Maurizio Vannoni, Gérard Poisson, and Pierre Vieyres  Abstract— This paper presents a new tele-operated roboticchain for real-time ultrasound image acquisition and medicaldiagnosis. This system has been developed in the frame of theMobile Tele-Echography Using an Ultralight Robot EuropeanProject. A light-weight six degrees-of-freedom serial robot, witha remote center of motion, has been specially designed for thisapplication. It holds and moves a real probe on a distant patientaccording to the expert gesture and permits an image acqui-sition using a standard ultrasound device. The combination of mechanical structure choice for the robot and dedicated controllaw, particularly nearby the singular configuration allows a goodpath following and a robotized gesture accuracy. The choice of compression techniques for image transmission enables a com-promise between flow and quality. These combined approaches,for robotics and image processing, enable the medical specialistto better control the remote ultrasound probe holder system andto receive stable and good quality ultrasound images to make adiagnosis via any type of communication link from terrestrialto satellite. Clinical tests have been performed since April 2003.They used both satellite or Integrated Services Digital Networklines with a theoretical bandwidth of 384 Kb/s. They showed thetele-echography system helped to identify 66% of lesions and 83%of symptomatic pathologies.  Index Terms— Image compression and filtering, medicalrobotics, mobile communication, remote center of motion, ultra-sound images. I. I NTRODUCTION R OBOTIZED telemedecine offers great medical advan-tages for medical experts who want to perform skilledactions, from an expert center for the benefit of a remotelylocated patient [1]–[3]. Robotic applications, such as forminimally invasive surgical interventions, help in reducingrisks of damaging delicate anatomical parts; for noninvasiveand invasive techniques such as guided biopsy, teleroboticsbrings accurate displacement of distant tools for the benefitof remotely located patients. Among all medical imagingtechniques, echography examination offers quick and reliablenoninvasive diagnosis for many pathological situations. Itallows a specialist to evaluate the degree of emergency fora patient. However, it is a skilled and “operator-dependent”technique. In areas with poor or reduced medical facilities, inisolated sites, in difficult accessible areas, and sometimes inemergency cases, there is not always an ultrasound specialist onhand to perform the first echography from which an emergency Manuscript received August 29, 2003; revised May 6, 2004. This work issupported by the European Commission under Contract IST-2001-32516.The authors are with Laborotoire Vision and Robotics, Universityof Orleans,Bourges 18020, France (e-mail: Object Identifier 10.1109/TITB.2004.840062 prediagnosis could be made. For these cases, a proposed alter-native to guarantee a reliable ultrasound examination is to use atele-operated robotized echography system. To guarantee a re-liable tele-examination and, therefore, diagnosis, it is importantfor the medical expert to forget about the distance between himand the patient. As a consequence, the remote control accuracyof the robot, the replica of the expert gestures by the robot,and the quality and flow of the received ultrasound imagesare correlated elements. They are needed by the expert in thefeedback control loop of the overall robotized tele-echographychain and necessary for the feasibility of the diagnosis.During the last five years, several laboratories have been in-volved with the development of robotic tele-echography appli-cations; for most of these, the challenge has been the combina-tionoftheroboticperformancesorthequalityofthetransmittedimage [4]–[7]. Most of these projects introduced parallel robotstofitthemedicalrequirementsofatele-echographyexaminationover the abdominal area of the remote patient. They are usuallyinstalled on the patient and cover the whole anatomical area.A different approach is proposed with the Mobile Tele-Echography Using an Ultralight Robot (OTELO) EuropeanProject. The main objectives of the OTELO system are toperform, in real time, a robotized echography examination on aremotely located patient, using any type of communication link from terrestrial to satellite. The main concern is to combine thedevelopment of a light-weight robot with appropriate imagecompression techniques to offer the medical specialist a per-forming tool for an efficient hand-to-eye coordination duringthe robotized tele-echography examination.Ultrasound images and probe-holder robot are two importantelements of the overall tele-echography chain. They are depen-dent from each other and they contribute to make the systemtransparent for the expert who is the major actor of the tele-operated control loop.The expert controls the remote probe holder robot by com-bining the one degree-of-freedom (DOF) hands-free input de-vice orientation and the received information. This informationincludes the ultrasound images and the position of the robot onthe patient’s body. When considering an ideal communicationlink, there is no time delay between the emission of the realprobe position data and the reception of the received image.Therefore, the real ultrasound plane position corresponds, atany time, to the desired one given by the input device held bythe expert. In a real tele-operated scenario, one should expecttime delay in the communication link combined with a non-negligible response time of the robotics and electronics system.Thereceivedultrasoundimagesdonotcorrespondanylongerto 1089-7771/$20.00 © 2005 IEEE  DELGORGE  et al. : TELE-OPERATED MOBILE ULTRASOUND SCANNER USING A LIGHT-WEIGHT ROBOT 51 Fig. 1. OTELO mobile robotized tele-echography chain. the desired ultrasound plane position given by the input device.The quality of the ultrasound image can also be altered by thechosen compression technique required by the communicationlink bandwidth. The resulting hand-to-eye coordination is thenhindered.Therefore, the development of the robotics system is stronglylinked with ultrasound image study when considering the per-formances of the tele-operated global chain. The general spec-ifications of the OTELO system are described in Section II.Sections III and IV present the requirements for the robot me-chanical structure and the control law management. Image pro-cessing techniques chosen for ultrasound images to be sent viathe available communication link are introduced in Section V.Clinical tests were performed in Barcelona Hospital and the re-sults are presented. Conclusions and future work are given inSection VII.II. OTELO S YSTEM  G ENERAL  A RCHITECTUREAND  S PECIFICATIONS The OTELO chain consists of three parts (Fig. 1).1) Theexpertstationwherethemedicalexpertremotelycon-trols, with a dedicated one-DOF hands-free input device(also called fictive probe) fitted with a six-DOF localiza-tion sensor, the positions and orientations of the distantultrasound probe located on the patient’s skin. The spe-cialistreceives,inalmostrealtimedependingontheavail-able bandwidth, the patient’s ultrasound images. The in-formation received at the expert station is integrated in anergonomic graphic user interface. Furthermore, a video-conferencing system between the two stations enables theexpert, the paramedic, and the patient to communicatewith each other during the robotized tele-echography ex-amination.2) The patient station is constituted of a light-weightsix-DOF serial robot that holds the ultrasound probeavailable at the patient site, a compact ultrasound device,a control and communication portable unit, and a video-conferencing system. A paramedic assists the patient. Hepositions the light-weight robot on an anatomic referencepoint on the patient’s skin according to the medicalexpert’s instructions and maintains it during the exam-ination. A strain gauge force sensor, embedded in theprobe holder, measures the contact force between the realprobe and the patient’s skin. The robot controller enables Fig. 2. Patient and probe positioning for abdominal and renal ultrasoundinvestigations. to limit this force to 20 N for the patient’s comfort andsafety. Medical images from the ultrasound device arecompressed before being sent to the expert. The patientstation is mobile, that is easily transportable, and can bequickly set up on the spot of use and connected to anyavailable communication infrastructure using transmis-sion control protocol/Internet protocol and unreliabledata protocol modes.3) Thecommunicationlink,dataexchangedbetweenthetwostations, expert and patient, include ultrasound images,robot controls, haptic information, ambient images, andaudio instructions. Most of the bandwidth is used for ul-trasound or ambient image transfer. Terrestrial links [e.g.,Integrated Services Digital Network (ISDN)], fixed andmobile satellite solutions, or 3G technologies (e.g., Uni-versal Mobile Telecommunications System) with variousbandwidth are currently being tested in the frame of theOTELO project.III. R OBOT  S PECIFICATIONS  A. Study of Probe Motion During a Standard Exam Standard ultrasound examinations have been analyzed inorder to quantify the positions, movements, and velocities of the probe used by ultrasound specialists.Fig.2showstherequiredprobepositioningandpositionsthatthe patient has to be set in for various ultrasound investigations:cardiac, abdominal, and renal. These are prime aspects takeninto account in the design of the serial six-DOF robot for theprobe positioning and for its handling during the robotized tele-echography.During these specification studies, interesting points havebeen reported: Continuous contact is kept between the probeand the skin, even when the probe is applied on ribs and irreg-ular abdominal skin. The contact force value varies from 5 to20 N. As shown in Fig. 2, the probe is most often held closeto the normal direction of the skin except in some cases, suchas in bladder investigation, where the probe can be tilted up to60 from the normal direction of the skin. Once the probe ispositioned on the area of interest, rotations, inclinations, and  52 IEEE TRANSACTIONS ON INFORMATION TECHNOLOGY IN BIOMEDICINE, VOL. 9, NO. 1, MARCH 2005 Fig. 3. Kinematics of the OTELO prototype.Fig. 4. OTELO prototype with its control portable unit and the compact PIEMEDICAL ultrasound device. small translations are performed around the chosen contactpoint with the skin.  B. Robot Mechanical Specifications According to the studied ultrasound probe motions, aportable compact six-DOF probe holder robot has been de-signed (Fig. 3) and manufactured with its control and powerportable unit (Fig. 4). It includes the following.1) Threerotations ( , , and , with their respectivero-tation axes , , and ) with a remote center of motionlocatedattheend-tipoftherealultrasoundprobeallowingtheprobetorotateinsideaconicalspace,withamaximum60 vertex angle.2) One translation ( ) of 4 cm along the probe axis direc-tion, defined by axis, allows the probe to be continu-ously in contact with the skin.3) Twotranslations( and )ofperpendicularaxesallowthe expert to adjust the position (5 5 cm) of the contactpoint above the organ being investigated. The two associ-ated directions and define a plane perpendicular tothe direction of rotation .4) The mechanical parameter defines the chosen fixedangle between and , and also between and .It allows a maximum tilt of the ultrasound probe of 45with respect to axis.A compact size and general purpose applications ultrasoundprobe as been chosen for first technical and clinical validations(see Section VI) performed in Barcelona Hospital Clinic sinceApril 2003.IV. R OBOTIC  C ONTROL  L AW Due to the serial robot kinematics structure, there is no needto develop and inverse the jacobian matrix as it is possible toobtain a simple analytic expression of the inverse geometricalmodel (IGM). The control laws are obtained by using the IGM[8]. However, the choice of geometrical parameters (e.g., axesranges) imposed by technical constraints (e.g., actuators, sen-sors, and mechanical components dimensions), and by the ne-cessity to have a mechanical equilibrium of mass (robot massover the center of the working space) bring two singular con-figurations inside the conical working space. One correspondsto the boundary configuration, as shown in Fig. 3. The othersingularity occurs when probe holder axes and (corre-sponding, respectively, to rotations and ) are collinear.This latter, corresponds to the probe in a vertical position andis called a wrist singularity or an “uncertain configuration” [9].In this case, large and fast movements of the axes are encoun-tered when moving near the singularity. This is a disturbing sit-uation for the medical expert as the received ultrasound planeis constantly changing. A study of kinematics solutions for therobot has shown that it is possible to design a robot withoutsingular configuration. However, the OTELO prototype bringsbetter performances in terms of weight and size which are oneof the prime user requirements.The first singularity has been overcome by using mechan-ical structure limitations and software limitations. For the wristsingularity, we developed a specific control law to accuratelyfollowthetrajectoriesinitsvicinityandtoavoidthespeedpeaksthatcannotbefollowedbytheactuators.Tocopewiththeselim-itations, two solutions are possible as follows:1) to trackexactly thedesired path inthe workspacebut witha reduced speed;2) to join the setpoints as quickly as possible for each artic-ulation without worrying about the path.The second possibility has been chosen as ultrasound exam-ination is noninvasive and as a few degrees of error in the ul-trasound plane orientation can be tolerated without hinderingthe organ search by the expert: As this expert is the main actorof the control loop of the tele-operated chain, he can contin-uously readjust the orientation of the input device accordingto the received image. We use the local redundancy of Axes 1and 3 to make the singular configuration no longer “uncertain.”When the robot reaches the singular configuration, movementis induced to only one of the two locally redundant articula-tion (either Axis 1 or 3). As Axis 1 presents a larger inertia,movements were given to Axis 3 which is controlled to performthe requested movement to obtain the closest orientation of theprobe to the desired one.  DELGORGE  et al. : TELE-OPERATED MOBILE ULTRASOUND SCANNER USING A LIGHT-WEIGHT ROBOT 53 Fig. 5. Reference and controlled vector for the ultrasound probe position. The IGM analysis shows an articular decoupling between theprobe position and rotation parameters which, however, impliesa specific computing process order. The implemented controllaw first computes the rotation parameters then defines the posi-tionparameters.Inthischosenmechanicalsix-DOFserialrobot,the wrist singularity only affects the rotation parameters and theoptimization process is, therefore, applied on these parameters.The desired unit vector of the probe axis is defined as thereference vector. and are the reference articular rotationsof the vector. is the rotation along the axis. isthe unit vector of the probe axis obtained with the optimizationprocess and is defined as the control vector (Fig. 5). It gives thereal controlled position of the ultrasound probe. , , andarethecontrolledarticularrotationsdeterminatedbytheprocesswhich will lead to the computation of followed by and .No tolerance error is considered on the probe nutation whichonly depends on Articulation 2: Therefore, equals . Thedetermination of the control articulation rotation is obtainedby using the following proposed filter:where is a nonlinear parameter varying with respect to thedistance to the singularity position [10], [11]. is the con-trolled value at , is the reference position at , andis the position of Articulation 1 at .Finally, the Articulation 3 control is chosen to obtain theorientation of the probe the closest to the referenced one.The results of the control law management nearby the singu-larity are shown without the controller in Fig. 6(a) and with thecontroller Fig. 6(b). Data are acquired from the robot articularsensors during a real robotic tele-echography examination. Thereference trajectory is generated by a medical ultrasound expertwho moves theone-DOF hands-free input device (fictiveprobe)in the vicinity of the wrist singularity.The articular movements managed with the implementedcontroller [Fig. 6(b)] are smoother than the ones obtainedwithout the controller. Notably, for the given input path, Artic-ulation 1 makes almost no move but is actually compensatedby Articulation 3 movements. Furthermore, Articulation 3movements remain smooth and with a smaller magnitude thanthe one observed without the singularity controller [Fig. 6(a)].This controller enables the specialist to move the remote realultrasound probe within the singularity area offering the de-livery to the expert station of a steady sequence of images even Fig. 6. Articular trajectory versus time, (a) without singularity processing,(b) with the singularity controller for the same reference trajectory. when in a normal position to the patient’s skin. In return, thankstotheperformancesofthesingularitycontrollertheremotecon-trol of the robot is improved and provides the specialist with areliable tele-echography chain to make a diagnosis.V. U LTRASOUND  I MAGE  P ROCESSING Echographic images are grabbed at the patient station usingany ultrasound device and sent to the expert station via a stan-dard communication link for which a good quality of service isensured.Two kinds of ultrasound image data can be transmitted to theexpert: video and still images. A video compression techniqueis used for the OTELO project in [12]. We focus, in this paper,on the compression of still images.Ultrasound image sequences can be transmitted followingtwo approaches. When the expert is searching for a specificorgan (liver, kidney, etc.), a high quality image may not be re-quired: Simple compression methods or lossy techniques can  54 IEEE TRANSACTIONS ON INFORMATION TECHNOLOGY IN BIOMEDICINE, VOL. 9, NO. 1, MARCH 2005 Fig. 7. Comparison between (a) srcinal image and (b) filtered image:PSNR       dB. be applied. When the organ of interest is found, it may be nec-essary to consider lossless compression techniques that wouldbring higher image quality to the expert. This lossless compres-sioncanbeappliedonthewholeimageoronaregionofinterest(R.O.I.). This choice also depends on the characteristics of thechosen communication link bandwidth.Inthispaper,wefirstevaluatethecontributionofaprefilteringstep in order to facilitate the compression. Then, we determinethe best dedicated compression technique to be applied on ul-trasound images. A comparative study among the most pop-ular compression techniques was performed to determine themethod that would give the optimal transmission time and thebest image quality for the ultrasound specialist. This compara-tive study uses statistical and psychovisual criteria.  A. Preprocessing of Ultrasound Images The goal of this survey is to improve the ultrasound image,in order to make its compression, and also its transmission,easier. Two different degradations can be detected in an image:the lack of contrast in ultrasound images and the presence of noise (speckle). When an ultrasound image has a low contrast,we give the medical expert the choice to adapt this parameter,because automatically changing contrast could affect the diag-nosis. On the other hand, it is well known that ultrasound im-ages are altered by speckle noise. This speckle is generated bythe back-scattered waves due to the heterogeneity of the humantissues. This noise can be reduced without affecting the medical judgment of the expert. We propose here to identify the param-eters of this speckle noise, and to apply a filtering method, bestadapted to these parameters, in order to correct the ultrasoundimage, and to improve the compression process.The system developed in [13] has identified in 73% of theimages a multiplicative noise. The estimation module [14]showed a standard deviation of 0.1. An acquired image islike with the denoisy image and a randomvariable with a unit mean and a standard deviation of 0.1. Withthese parameters, traditional filters, such as Wiener and Kuan,are used to denoise the ultrasound images.Fig. 7 shows a comparison between srcinal and denoisedimage.  B. Echographic Image Compression1) Experimental Protocol:  The survey was performed ona database composed of 20 ultrasound images of size 768576 similar to the one shown in Fig. 7(a). These images wereacquired by various ultrasound devices (Tringa, Sonosite, orESAOTE), then digitized thanks to a grabbing card (MatroxMeteor board, or Matrix MV-Delta). 2) Tested Compression Methods:  We compared the fol-lowing techniques: lossless image codings (Huffman, Arith-metic, Lempel Ziv; RLE and Fano codings) with lossy tech-niques (JPEG, JPEG-LS, and JPEG2000). For further details onthis study, one can see [15]. We show the comparison of thesetechniques by using some well-known statistical measures andsome psychovisual judgements. C. Statistical Comparative Results To compare the different compression methods and to makethe interest of filtering obvious, the image quality was first eval-uated with respect to four quality measures.1) The mean square error (MSE) given bymeasures the distortion brought by the compression. It isdefined by the mean of the square distances between eachpixel ( ) of the original image and each pixelof the rebuilt image . is the number of rows andthe number of columns.2) The peak signal-to-noise ratio (PSNR)represents an unbiased measure of the fidelity of the re-built image. More precisely, it represents the MSE, refer-enced with respect to the dynamics of the image in deci-bels. is the maximal intensity. The larger the PSNRis, the smaller the MSE gets, the better the rebuilt imagequality (that is to say faithful to the srcinal image).3) Compressionrate definesthesizeinbytesofthefinalimage over size in bytes of the srcinal image, expressedin percent.4) Coding times address lossless compression methods;compression and decompression times havebeen computed. For lossy methods, the global compres-sion time has been measured.The following results represent an average measure of each cri-terioncalculatedontheoriginalandrebuiltimages.Resultscon-cerning , , and have to be looked at in comparisonwith each other to appreciate the performance of each of thestudied techniques. 1) LosslessCase:  ItcanbeconcludedthattheRLEcodingisnotsuitedtoultrasoundimage,asits isthelargest(TableI).Fano et Huffman algorithms give comparable results in termsof , , and , with poor performances. The Adapta-tive Huffman method presents a compression rate of 54.57%(the final image size is about half of the srcinal one). The lastmethod, based on arithmetic coding, give the best compressionrate, but is associated with larger compression and decompres-sion times.The Adaptative Huffman method gives the best compromisebetweencompressionrateandcomputingtimes.JPEG-LSgives
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks