Добавил:
kiopkiopkiop18@yandex.ru t.me/Prokururor I Вовсе не секретарь, но почту проверяю Опубликованный материал нарушает ваши авторские права? Сообщите нам.
Вуз: Предмет: Файл:

Ординатура / Хирургия / Библиотека им академика М.И. Перельмана / Книга_3592_Библиотеки_им_академика_М_И_Перельмана

.pdf
Скачиваний:
0
Добавлен:
29.08.2026
Размер:
89 Мб
Скачать
Vascular and Intravascular Imaging Trends, Analysis, and Challenges, Volume 1
https://t.me/medicina_free
vessels. A series (typically 256) of ultrasound pulses are emitted and the consequent echoes are detected at different angular directions. Common IVUS systems employ 1D piezoelectric transducer arrays and generate 360° images on planes perpendic­ular to the blood ow. Conventional IVUS catheters used in the coronary arteries have a theoretical resolution of 19–39 μm (20–45 MHz) with >5 mm penetration.
While IVUS imaging is relatively affordable, based on disposable equipment alone, it is more expensive than the use of x-ray uoroscopy and is an invasive imaging procedure requiring the insertion and withdrawal of a catheter. However, the imaging modality itself is much safer, with no ionising radiation and no requirement for toxic contrast agents. Processing of the ultrasound images can be challenging but as automatic segmentation and characterisation of images are already available and implemented in commercial systems, this lends well to the use of this modality for intra-operative navigational purposes.
11.2.1.1 Registration to fluoroscopy
A common approach to the use of IVUS imaging for navigational purposes is through the registration of angiography to the series of collected IVUS images. A better understanding of the entire vascular network is obtained through the 3D angiography whilst the IVUS images provide the cross-sectional characterisation of the artery under consideration.
Accurate registration of IVUS images to angiography does require tracking of the catheter in the uoroscopic images. It is not the aim of this chapter to review this body of work, but many methods for the extraction of the catheter conguration have already been proposed. Fluoroscopy-based computer vision techniques have used optimal B-spline curve tting [3], adaptive spatio-temporal ltering [34], learning classification approaches [6] and dynamic optimisation [46], amongst others.
Wahle et al [43] presented an initial system for the fusion of biplane angiography with IVUS imaging. A 3D model of the vessel was rst obtained from the angiography data. While the IVUS images were collected, the biplane C-arm images were also recorded and a dynamic programming approach was used to detect the full catheter trajectory. A catheter model encapsulating bending, torsion, position, orientation and twisting was also incorporated to ensure that the IVUS images were oriented correctly on the catheter path. The authors also proposed an automatic statistics-based algorithm for the identication of correct 3D orientation of IVUS images during registration to the angiographic data [44]. Evans et al [12] took a similar approach, also using ECG-gated biplane uoroscopy to track the IVUS catheter and then using afne transformation matrices to fuse the IVUS images to the catheter positions. The benets of biplane uoroscopy are clearthe 3D positions of the catheters can be tracked from the images; however, biplane imaging systems are more expensive and their limited movement can affect the surgical workow and thus they are still relatively uncommon in hospitals.
Tu et al [40] approach the registration process for coronary procedures with distance mapping to a known, manually labelled position. They also skip the 3D reconstruction of the catheter trajectory, using only the centreline of the 3D vasculature. Distance mapping may not be suitable for more tortuous vasculature
11-3
Vascular and Intravascular Imaging Trends, Analysis, and Challenges, Volume 1
https://t.me/medicina_free
but is effective in more straightforward vessels. Manual point selection may be error prone but does facilitate the registration process. Wang et al [45] simplies the manual requirement of their approach by only having the operator dening a segment of clinical signicance. Any endovascular devices in this uoroscopic eld­of-viewwere then detected and tracked using a Haar-based probabilistic boosting tree classier and a Bayesian model, respectively.
11.2.1.2 Feature detection
Navigation using IVUS requires the identication of landmarks in the IVUS image sequence, both for registration purposes as well as localisation within the vascula­ture. Alberti et al [2] have developed a method based on classication of textural features and a multiscale stacked sequential learning scheme to automatically detect and measure bifurcations in coronary IVUS sequences. In addition to the bifurca­tions, the corresponding image frames, the angular orientation of the bifurcation and the extension are all identied.
Features can also include those from implanted devices, although this requires the IVUS imaging to be performed after the procedure. The fully automatic detection of stent struts in IVUS images was performed using a cascade of GentleBoost classiers in work by Rotger et al [30]; here, structural features of the stent were used to code the information of the different subregions of the struts. More recently from the same group, automatic strut detection is rst preceded by a stent shape estimation from the IVUS images using a supervised context-aware multi-class classication scheme [8]. The stent strut identication exploits both the estimated stent shape and the local appearance of the strut features.
11.2.1.3 Forward facing probes
Forward-looking IVUS image probes have been suggested and prototypes developed but to this date, these have not been made commercially available. In a navigational context, a forward facing IVUS catheter would be able to assist clinicians with the detection of vessel branches and vessel obstructions for guidewire advancement, without requiring the use of contrast agent enhanced x-ray uoroscopy.
Stephens et al [38] developed two forward-looking array designs for intracardiac use; these catheters were used within the heart simultaneously with an ablation catheter. While the initial designs were promising, the quality of the ultrasound images was less than that of commercially available catheters. In 2008, Volcano (San Diego, USA) acquired Novelis and their proprietary forward-looking IVUS (FLIVUS) technology with a view to integrating this into their existing platform. It has been suggested that they have been developing the technology and are looking to bring the product to market in the very near future but with the recent acquisition of Volcano by Philips (2014), it is now not clear if the product will appear.
11.2.2 OCT
Intravascular optical coherence tomography (OCT) imaging also requires the use of a contrast agent, which while safe for most of the population, may cause adverse
11-4
Vascular and Intravascular Imaging Trends, Analysis, and Challenges, Volume 1
https://t.me/medicina_free
reactions. Iodinated contrast agents, as also used in computed tomography angiography (CTA), can accumulate in the kidneys and lead to contrast-induced nephropathy [16]. The use of OCT has been compared to IVUS and studies have shown that the increased resolution of OCT can be used to better detect stent malposition. However, while there are many benets to OCT, the fact remains that it is less common in theatres than IVUS.
Unal et al [41] detected stent struts in OCT images for assessing the amount of in­stent restenosis post percutaneous coronary intervention. Their results show great promise for the detection of features (albeit features of already implanted stents) using OCT, with possible application to navigation.
11.2.3 Intravascular magnetic resonance imaging
Magnetic resonance (MR) imaging is a safe imaging modality without any ionising radiation and there has been an increasing amount of research into its use for endovascular procedural guidance via MR uoroscopy [31]. This is not strictly speaking an endovascular imaging modality, but MR is capable of real-time 3D localisation of the tip of the endovascular catheter and tissue characterisation. However, MR imaging for guidance of endovascular procedures is not currently in clinical use as there remain many research issues in the development of MR compatible endovascular devices and limitations to real-time MR imaging such as image resolution [18]. Also, the cost of MR is signicantly higher than that of other interventional imaging modalities.
11.2.4 Other sensing
Apart from imaging, a number of other sensing devices are also available for endovascular use. Investigations into their use for in vivo navigation have been performed, with initial results presented.
11.2.4.1 Electromagnetic sensing
The use of electromagnetic (EM) tracking with catheters has been explored extensively and is already in use in commercial systems for tracking cardiac catheters (e.g. Biosense-Webster CARTO). These can be used in conjunction with a preoperative scan of the vasculature, in which case registration is required. Fiducial markers on the skin may be used or natural features such as bifurcations or existing calcications in the vasculature.
Liu et al [23] have employed EM tracking to determine real-time catheter pose with respect to the surrounding vascular wall. The geometric information about the vascular walls was preoperatively obtained using MRI and registration between the preoperative image data and intra-operative vascular phantom were performed using a two stage registration scheme consisting of the iterative closest point (ICP) algorithm.
Wood et al [47] have investigated the use of EM sensors on many devices, including endovascular guidewires. Here, rather than simply attaching the EM sensors to the catheter, and thus affecting manipulation of the devices, they have
11-5
Vascular and Intravascular Imaging Trends, Analysis, and Challenges, Volume 1
https://t.me/medicina_free
created guidewires from the sensors, embedding them within the device itself. This leads to more natural manipulation of the device without added bulk. Validation was performed on phantoms. Also validated on silicone phantoms was work performed by Cochennec et al [9] using the Stealthstation
®
(Medtronic, USA), a commercially available clinical EM-based surgical navigation system, to guide catheter cannulation. They found that its use, though increasing the time for cannulations, reduced the use of x-ray uoroscopy and improved cannulation performance scores.
One of the most recent and advanced navigation systems available is the CustusX (SINTEF, Dept Medical Technology, Trondheim, Norway), originally developed for minimally invasive procedures. Again, guidewires were created with sensors embedded within. This still requires the connection of trailing wires to the EM control system and there are limitations as to the size of the eld created by the EM eld generator. The NDI Aurora (NDI, Waterloo, ON, Canada), used in this system, has a eld size of 0.125 square metres, smaller than that required to cover the length of the descending aorta in the average person. Some of this may be circumvented through the use of extra sensors used to roughly track the position of the anatomy, but any ferromagnetic device within or near to the electromagnetic eld will affect the eld and hence results. The authors have also reviewed the use of EM sensing for navigation, along with possible methods for error compensation, in endovascular aortic repair [39].
There have been clinical and pre-clinical investigations into the use of EM sensing in endovascular interventions. Abi-Jaoudeh et al [1] demonstrated the feasibility of inserting a thoracic stent graft with a navigation system incorporating both 3D images and EM sensing. Using the CustusX system, Manstad-Hulaas et al [24] has also shown successful catheterisations with fewer attempts required to insert the guidewire correctly than without the navigation system.
As is common with any system dependent on a static preoperative volumetric scan of the anatomy, however, motion and deformation affects the results. The aorta is not xed within the body and may deform as the patient moves. Likewise, the introduction of more rigid devices into the vasculature may cause deformation; a stiff guidewire may affect a straighteningof the vessel. This may affect registration results to the intra-operative EM coordinate system. Motion, in the form of cardiac pulsation or respiration, can also affect navigation results.
11.2.4.2 Shape sensing
Of recent interest is research into the use of bre-Bragg grating (FBG) for shape sensing. This is an optical sensor made of a multi-core bre, with changes in the optical length detected in each of the single cores. They are particularly sensitive to strain and temperature and thus have been used to sense changes in these in aerospace industries and industrial engineering, among others. Fibre optic shape sensing, an emerging technology based on bre-Bragg grating, has been demon­strated and patented by Luna Technologies [14], with results showing the 3D shape reconstruction of the entire length of a catheter. However, to date, this technology has not been incorporated into clinical devices.
11-6
Vascular and Intravascular Imaging Trends, Analysis, and Challenges, Volume 1
https://t.me/medicina_free
11.2.4.3 Catheter modelling
The shape of a catheter may also be reconstructed by modelling the mechanical properties of the device as well as its interaction with anatomical structures. This can provide an estimation of their current pose within the body [15, 20, 22, 37]. However, the use of mechanical modelling, such as nite element simulations, can be computationally expensive and difcult to implement for real-time processing.
The major advantage of these approaches, however, is the ability to provide a prediction on the effect of the catheter motion with respect to the vasculature. This can be particularly useful in situations where the interventionist wants to avoid particular areas of the vessel or assess beforehand the effect of a certain procedure. In addition, these models can be used to simulate the procedure and create a training environment for physicians.
11.3 IVUS for navigation
It is clear that t here is a need for an improved navigation system for endovascular interventions and there is scope to achieve this through the use of intravascular imaging. To this end, we have proposed two separate frameworks for navigation for endovascular procedures based on endovascular imaging. The rst method, combining IVUS imaging, catheter modelling and EM sensing, was initially proposed as part of the collaborative SCATh FP7 framework project [33]and then extended in the CASCADE FP7 framework project [7]. The second removes the need for explicit modelling and tracking, relying only on the IVUS imaging and a preoperative scan.
11.3.1 IVUS and EM sensing
Simultaneous localisation and mapping (SLAM) is a very popular approach in robotics [10] which aims to provide navigation capabilities to robots equipped with sensing capabilities (e.g. LiDAR, cameras, etc). In the SCATh project the concept of Endovascular SLAMwas introduced to provide a new navigation tool for catheter­based intervention based on information provided by EM and IVUS sensors integrated into a single catheter. The main motivation behind the proposed approach is to be able to obtain a 3D representation of the catheter shape and the vessel geometry which can be used by the surgeon to safely navigate the catheter within the vasculature without or with limited use of uoroscopy.
11.3.1.1 Catheter design
In the SCATh project, a customised catheter was developed where an IVUS sensor and multiple EM sensors were integrated, as shown in gure 11.1.
The IVUS system from Volcano Corporation with a Visions
®
PV 8.2 catheter was used. An NDI Aurora EM tracking system (NDI, Ontario, Canada) was used with their sensors (9 × 0.5 mm, 5 degrees of freedom (DOF) and 9 × 1.8 mm 6 DOF) integrated with the IVUS catheter. Six 5 DOF sensors along the catheter and one 6 DOF at the tip of the catheter were employed to obtain information about the catheter position and orientation with respect to the EM reference system. In order
11-7
Vascular and Intravascular Imaging Trends, Analysis, and Challenges, Volume 1
https://t.me/medicina_free
Figure 11.1. Custom made catheter, integrating IVUS and EM sensors, for the SCATh FP7 project.
to accurately track the catheter position and shape during the procedure, the 5 DOF sensors were placed at 125 mm intervals, with the rst 80 mm from the 6 DOF sensor at the distal end of the catheter, to obtain better accuracy near the tip. The 3D catheter shape within the vessel could be reconstructed and mapped onto a 3D preoperative model to provide an effective and intuitive visualisation during the catheter insertion.
In order to cope with registration errors, vessel motion and deformation, the IVUS images were processed to extract the relative position of the catheter tip with respect to the vessel lumen.
11.3.1.2 Catheter shape estimation and vessel geometry reconstruction
The fusion of this information in a probabilistic framework and the 3D visualisation of catheter and vessel were introduced as a novel navigation approach for endovascular intervention.
The catheter was dened by a nite number of nodes
where
=…Xxx x{, , , },
tn
12
i
θωϕ=xxyz(, , , , , )
represented the position and orientation at time t. The
T t
(11.1)
shape of the catheter was then estimated by a set of Catmull–Rom splines which were used to estimate the interconnection between the nodes.
In order to estimate the 3D position of the nodes a Kalman lter approach was used to integrate the information from the EM sensors and a physically based catheter simulation model based on real-time insertion length measures [11]. The catheter insertion simulation model provides the displacement of the nodes:
δδ δ=…Uxx x{, , , },
tn
12
T t
(11.2)
that can be used as a prediction model in the Kalman lter, which is dened as
11-8
Vascular and Intravascular Imaging Trends, Analysis, and Challenges, Volume 1
https://t.me/medicina_free
XAX BU , (11.3)
tt tt1
where A is an identity matrix and
is the normally distributed process noise.
ω
t
ω=++
By considering the EM sensors as the catheter nodes, the EM tracking system provided the measurement for the Kalman lter.
The localisation and shape estimation of the catheter through this probabilistic framework offers useful cues for catheter navigation:
1. The reliability of the catheter position estimate can be assessed considering the covariance of the estimate.
2. The prediction model provides information to estimate in advance the position of the catheter as a consequence of its insertion of a certain magnitude.
3. The combination of EM measurement and the catheter insertion model compensates for the possible inaccuracy of the single elements.
In order to provide an online mapof the vasculature during the insertion and accurately estimate the position of the catheter tip with respect to the vessel wall, the IVUS images were processed to extract the vessel lumen (gure 11.2). The IVUS processing algorithm is represented in gure 11.3. Since the images were captured from the Volcano IVUS system using a frame-grabber, a pre-processing step was necessary to remove the grid marks on the images. Once the marks were removed, the image was transformed to polar coordinates and a ood ll algorithm was applied to identify the lumen area. The contour was then identied by transforming the image back to Cartesian coordinate system, applying a threshold on the ood
Figure 11.2. Catheter shape reconstruction via EM sensors. Combining this with IVUS images allowed for the reconstruction of the vessel geometry.
11-9
Vascular and Intravascular Imaging Trends, Analysis, and Challenges, Volume 1
https://t.me/medicina_free
Figure 11.3. Workflow for IVUS image processing.
lled image and applying a contour detection algorithm. The centerline of the vessel
was estimated by tting an ellipse to the contour.
The 6 DOF EM sensor at the tip of the catheter was used to map the 2D IVUS­derived vessel lumen to a 3D representation. Each point of the contour in the 2D image coordinate system was then transformed to the 3D world coordinate system using the following transformations:
W
t
W
=TTT,
I
t
=XTX,
IEE
I
tWt
I
W
t
(11.4)
(11.5)
where T is the transformation matrix, I is the IVUS image coordinate system, E is the EM tracker coordinate system, W is the world coordinate system and
I
X
t
described in gure 11.4.
11.3.1.3 Results
The accuracy of the proposed algorithm for catheter shape estimation was validated in a 2D Plexiglas phantom created from an MRI scan of a patient aorta. A camera was used to record the catheter shape during the insertion and obtain the ground truth. The catheter is inserted into the aorta phantom with a constant velocity of
is
11-10
Vascular and Intravascular Imaging Trends, Analysis, and Challenges, Volume 1
https://t.me/medicina_free
Figure 11.4. IVUS image point coordinates.
Table 11.1. Average position error (mm) computed on 20 catheter poses. S1 indicates the sensor at the catheter
tip, S2 the sensor in the middle and S3 the sensor closer to the catheter distal end.
Sensor EM tracking Insertion model Fusion
S1 2.5 3.0 1.8 S2 2.9 3.4 2.1 S3 3.5 3.2 2.3
S
3.0 3.2 2.1
10 mm s−1by using an actuation mechanism. The results recorded in the experiments are shown in table 11.1.
The accuracy of the vessel shape reconstruction algorithm was also assessed in vivo. A porcine study was performed in The Intervention Centre, Rikshospitalet, Oslo, using a Volcano IVUS system and a Siemens Artis zeego interventional radiology system. To visualise the vessel in 3D within the animal, a contrast medium was also used during the dyna-CT scan. The animal was placed at different respiratory breath hold positions and catheter pullbacks were performed manually by the interventional radiologist, with the catheter withdrawal performed at a slow and consistent speed.
The errors of the generated centreline to the ground truth centreline can be seen in table 11.2 .
An ongoing EU FP7 project (CASCADE) currently focusing on transcatheter aortic valve implantation (TAVI) is currently taking the research forward, with improved IVUS processing and integration with the preoperative model. The simultaneous catheter and environment mapping (SCEM) [35] improves on the image processing and model reconstruction. A RANSAC operator was used to improve the ellipse tting in the absence of the entire contour and a full 3D model of the aortic arch has also been built. The research is being extended to dynamic models, with controlled simulations of the cardiac and respiratory cycles.
11-11
Vascular and Intravascular Imaging Trends, Analysis, and Challenges, Volume 1
https://t.me/medicina_free
Table 11.2. The errors for the in vivo experiments at three different breath hold positions.
Breath hold Mean error (mm) Std dev error (mm)
Max inhalation 3.3 1.6 In between 3.9 1.8 Max exhalation 4.1 1.6
11.3.2 Vessel navigation and retargeting
While this previous approach for intra-operative navigation was based on the use of IVUS imaging along with another localisation technology, its use is limited to the clinical adoption of that technology, in this case, electromagnetic tracking. For example, to date, the NDI Aurora is not currently approved for clinical use. In addition, combining the localisation technologyin this case, a wired EM sensor to the catheter or guidewire without constricting manoeuvrability is challenging.
To address these issues, we have recently proposed a framework to navigate through a patient-specic vasculature based only on IVUS imaging and a preoper­ative model [19]. The preoperative model can be derived from contrast-enhanced CT or MRI and provides a general map of the vasculature. This map is used to keep track of where the IVUS transducer is within the vasculature and is initialised with a single IVUS catheter pullback. This removes the need for intra-operative x-ray uoroscopy and the introduction of any separate localisation technology. We introduce the proposed framework here, showing initial results on phantom data.
11.3.2.1 Image processing and segmentation
No localisation technology was required, only a Volcano© Visions
®
PV 8.2 Phased­Array IVUS Imaging Catheter was used with a Volcano© s5imaging system (Volcano, San Diego, CA, USA). The VGA output of the imaging system was connected to an Epiphan VGA2Ethernet frame-grabber, with that output fed to a PC via an Ethernet cable. For validation of the results, a 6 DOF EM sensor was attached just proximal to the ultrasound transducer; this sensor was connected to an NDI Aurora electromagnetic tracking system (NDI, Waterloo, ON, Canada).
Pullback sequences of this IVUS catheter were obtained within a rigid Plexiglas aorta phantom (gure 11.5) produced by Materialise (Leuven, Belgium); the phantom was submerged within water heated to approximately 37 °C. For each sequence recorded, the IVUS catheter was inserted to the aortic root (using a guidewire) and a manual pullback was obtained. Approximately 1200 images were collected for each pullback sequence.
For registration between the EM tracking coordinate system and the CT coordinate system, a set of ve CT visible marks was mounted on the rigid phantom and the positions of these were collected in both systems. The ICP algorithm [5] was used to align these markers and provide the rigid transformation between the two coordinate systems.
11-12