Economy & Finance

8 pages
2 views

A hybrid approach to medical decision support systems: Combining feature selection, fuzzy weighted pre-processing and AIRS

of 8
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.
Share
Description
A hybrid approach to medical decision support systems: Combining feature selection, fuzzy weighted pre-processing and AIRS
Transcript
  computer methods and programs in biomedicine 88 (2007) 144–151  journal homepage: www.intl.elsevierhealth.com/journals/cmpb Development of an auxiliary system for the execution of  vascular catheter interventions with a reducedradiological risk; system description and first experimental results Giuseppe Placidi a , ∗ , Danilo Franchi a , Luca Marsili a , Pasquale Gallo b a INFM, c/o Department of Science and Biomedical Technologies, University of L’Aquila, Via Vetoio 10, 67100 Coppito, L’Aquila, Italy b Department of Internal Medicine and Public Health, University of L’Aquila, Delta 6, 67100 Coppito, L’Aquila, Italy a r t i c l e i n f o  Article history: Received 15 February 2007Received in revised form 3 July 2007Accepted 9 July 2007 Keywords: Vascular catheterizationPosition indicatorModel reconstruction3D RepresentationAngiographyX-ray reduction a b s t r a c t Vascular catheterization is a common procedure in clinical medicine. It is normally per-formed by a specialist using an X-ray fluoroscopic guide and contrast-media. In the presentpaper, an image-guided navigation system which indicates a path providing guidance tothe desired target inside the vascular tree is described with the aim of reducing the expo-sure of personnel and patients to X-rays during the catheterization procedure. A 3D modelof the patient vascular tree, reconstructed with data collected by an angiography beforestarting the intervention, is used as a guide map instead of fluoroscopic scans. An accuratespatial correspondence between the body of the patient and the 3D reconstructed vascularmodel is established and, by means of a position indicator installed over the catheter tip,the real-time position/orientation of the tip is indicated correctly.Thispaperdescribesthesystemandtheoperationalproceduresnecessarytousethepro-posed method efficiently during a catheter intervention. Preliminary experimental resultson a phantom are also reported.© 2007 Elsevier Ireland Ltd. All rights reserved. 1. Introduction Percutaneous vascular access and selective catheterization isa common procedure in clinical medicine.During catheter intervention, specialists (interventionalcardiologists, interventional radiologists, vascular surgeons,etc.), have to be guided by real-time imaging. Unfortunately,the usual imaging techniques, such as computed tomogra-phy (CT) and magnetic resonance imaging (MRI), cannot beused during catheter intervention because the scanners limitaccessibility to the patient and produce artefacts in real-time varying conditions. Specialists are, thus, constrained tomake use of fluoroscopic scans, with contrast-media, when ∗ Corresponding author . Tel.: +39 0862 433493; fax: +39 0862 433433.E-mail address: Giuseppe.placidi@cc.univaq.it (G. Placidi). performingtheseinterventions.Theexecutionofaninterven-tionalprocedureimpliesrisks[1–16],someofthemconnected to the accuracy of the catheter position indication procedure.The use of X-rays fluoroscopic scanning reduces the risk of damaging vessels due to catheter movements because it is ahigh spatial resolution and real-time technique, and allowseasy access to the patient by the operator. Conversely, thewide application of these procedures and the necessity torepeatthemseveraltimesleadstoseriousrisksofX-rayexpo-sure, both to the patient and the operators [17–24]. In fact, a coronary stenting involves an exposure to X-rays of about700 times that of a standard thoracic radiography [23]. Thus, side effects (leukaemia, X-ray induced tumours, etc.) become 0169-2607/$ – see front matter © 2007 Elsevier Ireland Ltd. All rights reserved.doi:10.1016/j.cmpb.2007.07.009  computer methods and programs in biomedicine 88 (2007) 144–151  145 10 times more likely than in normal conditions. For this rea-son, it is mandatory to reduce the exposure of patients andoperators, as suggested by international directives [25–27].The aim of reducing X-ray exposure can be realised byusing a 3D model of the patient vascular tree as a guide dur-ing the intervention. The 3D model is reconstructed by using the data collected by a diagnostic angiography such as com-puted tomography angiography (CTA), magnetic resonanceangiography (MRA), ultra sound (US) or other techniques,instead of X-ray continuous fluoroscopy [28–33]. Though CTA implies X-ray exposition, it is worth noting that—(a) a refer-enceangiographyhastobeperformedfordiagnosticpurposesbefore the intervention: if this is a CTA, it can be used during theinterventionalprocedurewithoutfurtherX-rayexposition;(b) the radiation dose is absorbed only by the patient, not byhealthpersonnel;(c)thepatientradiologicalexposureisabout30% lower than during continuous X-ray fluoroscopy [23]; (d) in CT, the radiation dose is protocol-based and determinis-tic: this is not true with continuous fluoroscopy. A hardwaresystem, capable of indicating the catheter position and ori-entation inside the previously reconstructed vascular model,is required. It is necessary to establish an accurate spatialcorrespondence between the real patient vascular tree andthe 3D reconstructed vascular model. Once the correspon-dence has been established, vascular intervention can becarried out using the 3D model as a map, instead of X-rayprojections.Inthefollowingsections,adescriptionofthesys-tem and preliminary experimental results on a phantom arereported. 2. System description The presented system consists of a position indicator mod-ule (hardware) and a series of methods (software) for dataacquisition, reconstruction, filtering and presentation. Theinterventionprocedurecanthusbeperformedusingareducednumber of X-ray scans, without any significant change to theintervention procedure itself.Theproposedsystem,whichfunctionsasreportedinFig.1,consists of the following subsystems:1. Acquisitionandreconstructionmethod,basedonthemostinformative selected projections, for 3D model reconstruc-tion by using a set of X-ray scans or other diagnostic tools.2. Registration method to align, both in position and inscale, the reconstructed 3D model with the patientbody (this step is necessary because the 3D modelacquisition/reconstruction are performed away from theinterventionsiteandatadifferenttimeandtocompensatefor patient motion during the procedure).3. Catheter position/orientation indication system. Thisapparatus is necessary to indicate the position/orientationof the catheter tip during the intervention procedure with-out the use of X-ray scans as a guide.4. Graphic presentation method for the 3D spatial rendering of the reconstructed model and of the real-time move-ments of the catheter tip inside it (i.e. inside the patientbody in the same position).The system is completed by the procedure to be followedwhen using the proposed system during a catheter interven-tion.The reduction in the radiological risk to the patient andthe personnel involved is ensured both by using the adap-tive acquisition method [34,35], which reduces the number of projections necessary to collect the basal angiography bycollecting the most informative projections (in the case of X-ray based techniques), and by using the position/orientationindicator system. The accurate representation of the patientvascular tree reconstruction is guaranteed by reconstructionalgorithms from an incomplete set of projections and by fil-tering and interpolation methods [34–37].The present method ensures the absolute independenceof the diagnostic tools which are used to collect the angiog-raphy. Although the method requires the angiography to beperformed before the intervention (the virtual model drivesthe specialist), there is the possibility of using different angio-graphic methods without loss of efficacy. For example, if MRA is used instead of the classical, X-ray based, angiogra-phy, a further reduction of the radiological risk is possible.In what follows, we will use CTA to obtain the 3D modelfor the proposed experimental results. CTA leads to a spatialaccuracy which cannot actually be obtained by using other,non-invasive, imaging modalities such as MRA or US. In fact,MRA suffers from magnetic field in-homogeneities which candisturb spatial localization [38] and US is often affected byimproper probe calibration and its precision depends on thedepth of the image zone from the transducer [33,39]. 2.1. Acquisition and reconstruction of the 3D virtualvascular model The procedure for the acquisition/reconstruction of the 3Dvirtual vascular model can be summarised by the following steps:(1) Adaptive acquisition of a limited set of projections (themost informative).(2) Separation of vascular information from the background(signal coming from other organs or due to noise) throughthe segmentation of the projection images.(3) 3D reconstruction through iterative methods allowing both the collection of vascular spatial position and theirsignal intensity.Step 1 is realised by using modified versions of algorithmsdeveloped in the field of magnetic resonance tomography andappliedforacquisitiontimereduction,i.e.magneticresonanceangiography [34–36]. Unlike previously developed methods, theirvariantsallowtheanalysisofsome2Dprojections(X-rayorother)throughthecalculationofananalyticalfunctionaftera filtering process has been applied to distinguish the use-ful vascular signals from the background (from other organs,extraneous signals and noise). The segmentation algorithmused is based on the cross-correlation between the 2D projec-tion to be segmented and an anisotropic 2D Gaussian filter,at different angular orientations. After the correlation, a listof correlation values, one for a given angular orientation of the filter, corresponds to each pixel. Vessels are considered to  146  computer methods and programs in biomedicine 88 (2007) 144–151 Fig. 1 – Representation of system modules and procedure flow diagram to execute interventional vascular catheterization. be present on pixels where the maximum correlation value isaboveagiventhreshold.Adetaileddescriptionoftheproposedmethodwillbegivenelsewhere.Thereconstructionalgorithmuses,asa-prioriinformation,thepseudo-circularshapeofthevessel transversal section and the fact that the vessels are“filled”objects,i.e.theirtransversalsectioncannotbeannular(i.e. empty in the centre). This allows an accurate 3D ves-sel tree reconstruction with a reduced number of projections,those having maximum information content, thus reducing X-rayexposurewhenusingCTorstandardX-rayscans.WhenMRA is used, the application of this method is not strictlynecessary (perhaps it can be used to reduce the acquisitiontime). 2.2. Registration Method A registration method between the reconstructed 3D ves-sel tree model and the patient body is necessary in orderto ensure an accurate correspondence in position, orienta-tion and scale. This process can be realised in two differentways. In the first, the procedure is applied when the inter-vention starts: after the specialist inserts the catheter intothe input vessel, he immediately collects an X-ray scan withcontrast media at a given angular direction. At this point, theregistration method aligns the 3D model with the collected X-ray scan. The registration procedure consists of the following steps:(1) A numerical projection of the 3D reconstructed numericalmodel is calculated in the same direction as that of theX-ray scan collected by the specialist.(2) The numerical projection from the 3D model is registeredwiththerealX-rayscanandtranslation;rotationandscal-ing factor are calculated between the two images.(3) If the previously calculated values fall below a giventhreshold, then go to step 6.(4) The spatial transformation is applied to the 3D model.(5) Go to step 1.(6) Stop(the3Dmodeliscorrectly“aligned”withthevasculartree of the patient).Theregistrationalgorithmtobeusedinstep2canbeoneof the well-known methods used for medical images [37,40,41].Inparticular,aFouriermethod,basedonthecross-correlationmaximisation between the numerical 2D projection extractedby the 3D model and the real X-ray scan, is found to be thefaster and simpler method to calculate the optimal transla-tion, rotation and scaling factors. The alignment method iscompletelyautomaticandisexecutedinalmostreal-time.Thedescribed procedure can be easily implemented if traditionalcatheterization instrumentation and methods are used (X-rayscansandX-rayangiography).Itsexecutioncanbeinitiatedbytheoperatorthroughagivencommand(thiscommandcanbethe same as that used to collect the X-ray scan). Its executioncan also be repeated during the interventional procedure if required (this can occur if the alignment is lost, for exampledue to patient movements).Nevertheless, this constitutes a drawback because theoperator has to repeat the alignment when he realizes thatthecathetertipisfallingoutsidethemodel:thiscanobviouslyoccur also in the case of vessel rupture.Alternatively,toavoidthepreviouslimitation,spatial,scaleand orientation alignment (registration) can be performed byusing points of reference if MRA or CT 3D-model collection isused.Infact,asetof4–5pointscanbepreliminarilychosenaspointsofreferenceandfixedbothonthereconstructedmodeland on the patient body (these points can be both superfi-cial anatomical structures or structures artificially anchoredon the patient body before the angiography and maintainedin position during the catheterization intervention). Thesepoints can be used to perform a 3D transformation in orderto register the 3D model points with the corresponding pointson the patient body, i.e. to register the 3D reconstructed vas-cularmodelwiththecorrespondingpatientvasculartree.Thetransformation describes the motion of a non-rigid objectand corrects scaling errors as well as rotation and translation  computer methods and programs in biomedicine 88 (2007) 144–151  147 errors.Thislastprocedurehastheadvantageoffurtherreduc-ing the radiological exposure but can be applied efficientlywhenMRAorCTareusedduetotheircapacitytocollectinfor-mation about other structures, as well as vessels. This lastalignmentandregistrationcontrolprocedureisexecutedcon-tinuouslybythesystem:ifthepatientmoves,itisalsopossibleto maintain the correct spatial correspondence between the3D model and the patient body, independently of the opera-tor.Conversely,thisprocedurerequiresthespecialisttoacceptmodifications to the protocol he currently uses for catheterinterventions.Forthedescribedexperimentalresults,weusedthis last registration procedure, though the described systemsupports both modalities. 2.3. Position/orientation indication hardware The position/orientation indication system is similar to a GPS(Global Positioning System) which indicates the current posi-tion of an object on the earth’s surface by using satellites anda receiver positioned on the object. An existing apparatus foruse in angiography applications is Microbird ® , produced byAscension Technology [42], which uses a magnetic field mea- surement system installed on the tip of the catheter. A staticmagnetic field is produced by coils positioned externally tothe patient body. The acquisition system consists of a probewhich collects the magnetic field components and transmitsthem to a console through a cable (which also serves as aline feed) inside the catheter (Microbird ® can drive up to fourprobes of the same type). A better choice would be to usean innovative version of the system proposed elsewhere [43]:in this case, the source of the static magnetic field is posi-tioned on the tip of the catheter and the magnetic field probesare positioned externally to the patient body. Thus no cableneedstobeinsertedintothecatheterlumen.Forthedescribedexperimental results, we used Microbird ® with one installedprobe. 2.4. Rendering software for real-time presentation This method uses 3D graphical presentation of the recon-structed model. The volumetric rendering is one of theinnovative representation techniques used in medicine, inparticular in medical imaging and radiology. During vascu-lar catheter interventions, for example, the operator has toobserve a volume from different points of view, by frequentlyrotating, zooming and shifting the object to verify the currentpositionofthecatheterinsideit.Inthepresentedsystem,thisgoal is attained by implementing a dedicated computer pro-gram written in C++, by using VTK libraries [44]. The proposed software implements the visualisation of both the 3D modeland the real-time position of the object inside it.Moreover, the software implements other methods, suchas:(a) calculation and correction of the displacement betweenthe 3D model and the real body position;(b) verification in real-time of the catheter position so as toperform a realignment when its position differs from thatof the vessel position (i.e. when the catheter tip is indi-cated as being outside the vessel).The presentation and visualization method consists of thefollowing steps:(1) Graphic representation of the 3D model.(2) Alignment of the 3D model with the patient body vasculartree.(3) Position/orientation calculation of the catheter inside it.(4) Graphicrepresentationofthecurrentposition/orientationof the catheter.(5) If the current catheter position is internal to the vasculartree,(5a) Go to step 3;Else,(5b) Go to step 2. 2.5. Operational procedure The catheterization procedure, modified to reduce X-rayexposure, is realised by the utilisation of an integrated hard-ware/software support system. The hardware part consistsof Microbird ® with one sensor. It serves to check the spatialposition and orientation of the catheter tip and to visualizeit in the virtual vessel representation of the patient vascularanatomy. The virtual vascular model can be reconstructed byan angiography performed before starting the intervention:in this case, the angiography can also be collected by CTA,MRA, US or other scanners, also with triggering by dynamicevents (such as respiration or beating heart), without limi-tations. By using adaptive techniques, the execution of theangiography can also be optimised to reduce X-ray exposure.The 3D reconstruction of the patient vasculature can be usedby the operator as a map during the procedure.The classical interventional procedure can be modified toinclude the previously described systems as follows (see alsoFig. 1):(1) A set of reference points is defined on the patient body.(2) A basal angiography is performed (X-ray, CTA, MRA, orother).(3) The 3D vascular model is reconstructed.(4) The correct correspondence between the patient vas-cular tree and the 3D vascular model is established,using the registration method on reference points, andis continuously and automatically controlled during theintervention.(5) MicroBird ® is used to drive the catheter inside the patientbody using the 3D model as a map, rendered by the visu-alization software.It should be noted that step 1 requires extra preliminarytime in order to define reference points (anatomic regions orartificial objects anchored on the patient body).When the catheter reaches the goal site, the procedure canbeconventionallyconducted,byusingclassicalX-rayscans,inorder to obtain a greater precision and to verify the interven-tionalprogress.Aspreviouslydescribed,theuseofX-rayscansis limited to the critical phase of the intervention, resulting ina great reduction of the radiological risk both to the patientand to health personnel.  148  computer methods and programs in biomedicine 88 (2007) 144–151 Fig. 2 – Vascular model phantom, composed of empty glassand rubber tubes, used to test the proposed auxiliarycatheterization system. Wood supports, of different woodtype, are visible. On the supports, holes (2mm indiameters) have been made to be used as points of reference and some of them are also visible and circled withdashed curves. The photo also shows a metre ruler restingon the phantom to indicate its length (about 750mm). 3. Results and discussion In order to test the system, we simulated a vascular catheterintervention by using a phantom roughly representing a ves-sel tree, shown in Fig. 2, composed of empty rubber and glass tubes (the minimum useful diameter was 2mm, the max-imum was 2.5cm) anchored to wood supports with plasticglue. Several holes were drilled in the wood supports to actas reference points (in Figs. 2 and 3 these are indicated with dashed circles). The phantom was inserted into a commercialHitachi CT scanner and a complete spiral scan was performed Fig. 3 – 3D model, reconstructed by using the CT spiral scandata, rendered by using the proposed presentationsoftware. The phantom is shown from a point of view fromwhich a lot of points used as references are visible (they areindicated by dashed curves). (a set of 250 transversal 512 × 512 images, FOV 370mm, dis-tance between consecutive images: 3mm) and a 3D modelreconstruction was obtained, reported in Fig. 3 (A and B rep- resent different points of view). After data collection and 3Dmodel reconstruction, the phantom was moved to the opera-tive bed and one of us, specialised in interventional radiology,conducted the simulated intervention. First of all, the spe-cialist chose a series of 5 points of reference both on thereconstructed model and on the real phantom (these pointsare clearly visible on the reconstructed 3D model reported inFig.3).UsingsoftwareandMicrobird ® ,thefollowingprocedurewas performed:(1) For each point of reference:(a) The tip of the Microbird ® probe was positioned on thedesired target point of the real sample.(b) A3Dcursorwaspositionedonthecorrespondingposi-tion of the 3D reconstructed model.(c) These two positions were registered (locked).(2) The set of points of reference was used to fill the transfor-mation matrix which was applied to the 3D model to alignit to the real object.The alignment procedure, a hand made procedure for thisexample, took about 1min. The phantom was enclosed in abox so that the specialist could not see it and the interven-tional procedure started. By using visualization and operationsoftware, it was possible to reproduce and visualise the 3Dmodel and the real-time movements of the catheter tip insideit with a resolution of at least 1mm, both from externaland internal views, see Fig. 4. The measured resolution was about 1mm: this was the dimension of the lower structureexplored in the proposed phantom. The software used allowsthe importation of numerous data formats; to reconstructthe interpolated 3D model structure, to filter and also ren-der it in a transparent form (see Fig. 4A), and to calculate and apply a 3D spatial transformation in order to compen-sate for misalignments between the numerical model andthe real body. The specialist was able to explore internally(see Fig. 4B) the whole phantom and to reach, with the above precision, all of the desired locations with a 7 frenchscatheter. The experiment took about 10min, even though the3D model was realigned twice with the phantom, after theinitial alignment, due to movements of the phantom during the catheter intervention (the phantom was not fixed in posi-tion and it moved sometimes when the specialist moved thecatheter). The efficacy of the alignment procedure was thusverified.Some choices made during the execution of the describedprocedure need further explanation:(1) The transformation can suffer from two errors: onedue to the approximation of the 3D model acqui-sition/reconstruction technique; the other due to notpositioning the sensor precisely on the points chosen asreferences. Because of this, the number of points of refer-ence chosen is greater than necessary (in ideal conditionsthreepointswillsuffice)andthetransformationisappliedinaleastsquaressense.Thechoiceofagreaternumberof points than necessary will reduce these errors, especially
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