Evaluation of guidewire path reproducibility

Abstract

The number of minimally invasive vascular interventions is increasing. In these interventions, a variety of devices are directed to and placed at the site of intervention. The device used in almost all of these interventions is the guidewire, acting as a monorail for all devices which are delivered to the intervention site. However, even with the guidewire in place, clinicians still experience difficulties during the interventions. As a first step toward understanding these difficulties and facilitating guidewire and device guidance, we have investigated the reproducibility of the final paths of the guidewire in vessel phantom models on different factors: user, materials and geometry. Three vessel phantoms (vessel diameters ∼4 mm) were constructed having tortuousity similar to the internal carotid artery from silicon tubing and encased in Sylgard elastomer. Several trained users repeatedly passed two guidewires of different flexibility through the phantoms under pulsatile flow conditions. After the guidewire had been placed, rotational c-arm image sequences were acquired (9 in. II mode, 0.185 mm pixel size), and the phantom and guidewire were reconstructed (512³, 0.288 mm voxel size). The reconstructed volumes were aligned. The centerlines of the guidewire and the phantom vessel were then determined using region-growing techniques. Guidewire paths appear similar across users but not across materials. The average root mean square difference of the repeated placement was 0.17±0.02 mm (plastic-coated guidewire), 0.73±0.55 mm (steel guidewire) and 1.15±0.65 mm (steel versus plastic-coated). For a given guidewire, these results indicate that the guidewire path is relatively reproducible in shape and position.

INTRODUCTION

Cerebrovascular disease ranks third in cause of death, resulting in more than 160 000 deaths in the United States annually.¹ Initially, the standard approach to cerebrovascular interventions was invasive surgery. However, with the introduction of the Seldinger technique,² the site of disease can be accessed by using the vasculature itself as a pathway. In minimal vascular interventions, a remote artery, e.g., the femoral artery, is punctured, and a catheter is placed in the artery and brought to the interventional site of interest under the fluoroscopic guidance, often with the assistance of a two-dimensional (2D) roadmap. To facilitate this process, a guidewire (metal∕plastic-coated wire between 0.2 and 0.6 mm in diameter) is inserted through the catheter and is advanced ahead to guide the catheter through the vasculature. The guidewire is semiopaque to x rays and so can be seen in fluoroscopy. The guidewire is usually guided to and then beyond to the site of intervention. The catheter and the interventional device(s), e.g., stents, balloons, or coils, are brought to the site of intervention by moving them along the guidewire, i.e., the guidewire functions as a monorail for these devices.

Due to the tortuousity of the vessel and the relative stiffness of the guidewire and interventional devices, guiding the guidewire and the devices to the site of intervention can be difficult even for trained interventionalists. Improper forces on the vessel during guidewire or device placement can result in vessel perforation or dissection. The technical success of the procedure can depend strongly on the experience of the surgeon as well as the tortuousity of the access path, with up to 10% of interventions resulting in complications.^{3, 4, 5, 6} To improve the physicians’ technical skills, endovascular intervention simulators have been developed,^{7, 8, 9, 10, 11, 12, 13} providing some experience, albeit artificial, prior to the actual surgeries. During surgery, the physician must rely on the acquired images and his sense of touch to bring the guidewire and then the devices to the region of interest in the patient’s vasculature without harm. Because the guidewire plays such an important guidance role, assessment of the guidewire path prior to the actual surgery may be useful for planning minimally invasive procedures.

Clinicians (neurosurgeons and coronary interventionalists) have informed us that the guidewire path seems to be relatively reproducible in vivo. They also inform us that they will use a more rigid guidewire to straighten a vessel on occasion. Changing the vessel path can cause complications in the procedure by possibly hemorrhaging the vessel wall, pronouncing vessel tortuosity, causing significant spasm which reduces the blood flow and vessel plaque release.⁴ This can occur especially in symptomatic and older patients, which usually have weakened vessel properties (flexibility, distensability) and∕or high calcification rates. To model this vascular∕guidewire interaction it is necessary to know the positions at which the guidewire is touching the vessel wall, because only at these positions will force transfer occur and initiate the vascular deformation. To move forward in developing methods capable of calculating the approximate guidewire path and subsequent calculation of vessel deformation, a tool enabling the surgeon to decide on guidewire stiffness to achieve a desired reshaping of the vasculature and simultaneous assessment of complication risk, it is necessary as a first step to investigate the path achieved by the guidewire before any deformation occurs. To investigate this pre-deformation position we constructed rigid vascular phantoms and repeatedly inserted guidewires of different materials under conditions mimicking the circulatory system.

In preliminary studies¹⁴ using conventional biplane angiography, we found that the path of the guidewire was fairly reproducible. Therefore, in the current study, we obtained rotational angiography data sets from which we reconstructed the three-dimensional (3D) vessel and guidewire paths to investigate more completely the reproducibility of the guidewire path.

METHODS

Vessel phantoms with geometries similar to the internal carotid were constructed. Guidewires were passed through the phantom twice by experienced users. After placement of the guidewire, rotational c-arm cone beam image sequences were obtained, and the 3D phantom-guidewire volume was reconstructed. The guidewire and the vessel lumen were segmented using region growing, and the centerlines and vessel diameters were determined. The multiple guidewire paths were then compared.

Phantom models of the human internal carotid artery

Several groups have proposed methods to construct vessel phantoms.^{15, 16, 17, 18, 19, 20, 21, 22, 23} Although they have been developed for different applications, there are basically three phantom designs: walled,^{15, 16, 17, 18, 19, 20} nonwalled^{21, 22} and real-vessel phantoms.²³ In each design, the vessel structure is encased in a tissue-mimicking material, which usually has a low x-ray attenuation coefficient. In this study a walled-phantom design was applied.²⁴ The encasing material was Sylgard (Dow Corning, Midland, USA), a silicon-based compound. It has a low x-ray attenuation coefficient, is liquid prior to curing, and is transparent and rigid after curing. The vessel wall was constructed from a polyethylene tube. The tube provided low surface friction, was easily deformable, and had a low x-ray attenuation coefficient. The inner diameter of the tubing was chosen to be 3.65 mm and 4.5 mm following the results published by Bharadvaj et al.²⁵ for the diameter of the internal carotid artery. To achieve a shape similar to those of the internal carotid artery, we reconstructed the 3D centerlines of a number of carotids from multiview angiograms²⁶ and recorded their shape. Combinations of wires were shaped to resemble the features seen in the renderings of these 3D centerlines. These wires were then passed through the polyethylene tubes, which took on the geometry of the wires. The tubes containing the wires were placed in a 4-cm-diam cylinder. The cylinders were filled with Sylgard and hard cured. After curing, the wires were removed from the polyethylene tubing, and the cylindrical casing was removed from the Sylgard. Three phantoms were constructed using this method, an example is shown in Fig. 1.

Figure 1.

Vessel phantom. The tube length in the phantom is 143 mm, and the lumen diameter is 3.65 mm.

Experimental setup and data acquisition

The phantoms were fixed on the patient bed of a Toshiba Infinix Angiography system (Toshiba Medical Systems Corporation, Tokyo, Japan) to prevent movement of the phantom during guidewire insertion. To allow simultaneous fluid flow through the phantom and guidewire insertion, a valve system was designed and attached to each phantom prior to image acquisition. A pulsatile flow pump was used to generate flow which approximated that inside a human carotid artery. Blood was simulated by a 30∕70 glycerin∕water mixture. Two guidewires of different flexibility were used, one consisting of twisted stainless steel (called here steel) and the other consisting of plastic-coated steel wire (called here plastic-coated), each having a diameter of 0.36 mm. The elastic potential energy, also known as Young’s Modulus, for the two materials were 200 GPa (stainless steel) and 2.5 GPa (plastic coated). Before guidewire insertion, rotational sequences (FOV 9 in., SID 110 cm, SOD 80 cm, 2° intervals, 1024×1024 pixels) of the air-filled and of the contrast-filled vessel phantom were obtained and reconstructed using vendor-supplied software using a variation of the filtered backprojection as described by Feldkamp et al.;²⁷ the system was calibrated following the procedure recommended by the vendor.²⁸ Five users, experienced in the use of guidewires, passed the guidewire through the entire vessel phantom twice. After each insertion of the guidewire, rotational angiographic sequences were obtained, and the phantom and guidewire were reconstructed. Figure 2 visualizes the different vessel shapes embedded in the phantoms.

Figure 2.

From left to right: Phantom A (lumen length 23 cm, lumen diameter 4.5 mm), Phantom B (lumen length 19.1 cm, lumen diameter 4.5 mm), Phantom C (lumen length 15.6 cm, lumen diameter 3.6 mm).

3D vessel∕guidewire centerline extraction

The reconstructed 3D data consisted of 512³×8 bit volume, with a 0.288 mm voxel size. The voxel values ranged from 45 to 60 (Sylgard), 10 to 20 (blood mimic), 25 to 35 (patient table), 120 to 255 (guidewire) and 255 (contrast filled lumen). The 3D vessel lumen and guidewire were segmented in the images using thresholded region growing. While automatic threshold selection methods exist,²⁹ we decided to set the growing thresholds (for the vessel lumen, and for the guidewire) based on evaluation of the histograms and kept them constant for all data sets. The centerlines of both vessel and guidewire were determined using region growing accompanied by a wavefront-based technique.³⁰ Region growing was initiated near the most proximal or distal vessel∕guidewire centerline point. The seed point was chosen to be geometrically inside the Sylgard enclosed vessel region and region growing was stopped while inside the Sylgard enclosed vessel area near the top of the phantom. For the first seed, every neighboring voxel (eight-point connectivity) above the threshold was identified, added on a seed list, and marked as visited. This concluded the first iteration. In the following iterations, the most recent entries into the seed list were used as seeds for the next iteration of region growing. The center of mass of all grown points for a given iteration of region growing was determined by averaging to provide a local vessel∕guidewire centerline point with subvoxel precision. The region-growing and concurrent centerline determination continues until no voxels are found which satisfy the region-growing criteria.

Alignment of volumes

To correct for the variabilities arising from gantry or residual motion of the phantom, all reconstructed 3D phantom data (containing a guidewire) were aligned to the reconstructed 3D air-filled vessel phantom data based on the voxels corresponding to the Sylgard. Histograms of the voxel values were obtained of each 3D data set. The regions in the histograms corresponding to air, table, and phantom were identified. The voxels corresponding to the Sylgard portions of the phantom were identified as those having values between two thresholds set on either side of the histogram peak corresponding to the Sylgard. We chose to use the Sylgard for registration of the volumes to eliminate bias from the guidewire. The data were converted to a binary volume by threshold segmentation with the Sylgard voxels having a value of one; the voxel values of the other structures were set to zero. The centers of mass of the segmented phantom data were then calculated, and the 3D data containing the guidewire were translated by the differences in the centers of mass. With the centers of mass aligned, the rotation relating the two volumes was determined using a downhill Simplex technique, with an objective function given by the root mean square (RMS) difference of the intensities of the voxels in the two volumes, i.e.,

$$\mathrm{RMS}=\sqrt{\frac{{\sum }_{i=0}^N{(p(j,k,l)-p_{i,o,m,n}^{\prime }(j,k,l))}^2}{N}},$$ (1)

where p(j,k,l) represents the voxel value at location (j,k,l) in the air-filled vessel phantom, $p_{i,o,m,n}^{\prime }(j,k,l)$ represents the voxel value at location (j,k,l) in the phantom data rotated by θ for phantom i, material o, containing the guidewire inserted by user m the nth time. The determined rotation and translation were then applied to the corresponding 3D guidewire centerline, GC_i,o,m,n data. [Our implementation was done in C++ using the GNU Scientific Library (GSL)³¹].

Evaluations

The sensitivity of the phantom alignment to the thresholds used in region growing was assessed by comparing rotations and translations [using Eqs. 2, 3, respectively] obtained when the thresholds were extended and reduced by two intensity levels on either side (the range was chosen to prevent the inclusion of nonSylgard volume structures).

$$\varphi ={\cos }^{-1}\left(\frac{\mathrm{Trace}\left(R_A^T{R}_B\right)-1}{2}\right),$$ (2)

$${\vec{T}}_C={\vec{T}}_A-{\vec{T}}_B,$$ (3)

where R_A and R_B are rotations and T_A and T_B are translations determined with data segmented using sets A and B.

For point-to-point measures, the correspondence between points on the centerlines was first determined as follows. The guidewire centerlines g were interpolated using a B spline using a sampling distance of 0.04 mm. For each extracted vessel centerline point in the phantom data, the plane perpendicular to the centerline tangent at that point was determined. The guidewire centerline point lying in that plane was then determined for each case (user, insertion, and material) of that phantom. Thus, for each vessel centerline point, we created a set of corresponding guidewire centerline points whose 3D locations could be compared.

With the correspondences established, the variation of the guidewire about the mean location was determined. The mean guidewire path $\overline{g}$ was first calculated for each phantom i for guidewire material o for user m, averaged over insertions n using

$${\overline{g}}_{i,o,m}\left(j\right)=\frac{1}{N}\sum _{n=1}^N{g}_{i,o,m,n}\left(j\right),$$ (4)

where j corresponds to the guidewire centerline point index, and N is the number of insertions over which the average is performed. Subsequently, the RMS distance between the guidewire paths for each phantom, material, and user were calculated at each centerline point and for the complete guidewire centerline as:

$${\mathrm{RMS}}_{i,o,m}\left(j\right)=\sqrt{\frac{1}{N}\sum _{n=1}^N{(g_{i,o,m,n}\left(j\right)-{\overline{g}}_{i,o,m}\left(j\right))}^2},$$ (5)

$${\mathrm{RMS}}_{i,o,m}=\sqrt{\frac{1}{J_{i,o,m}}\sum _{j=1}^J{\left({\mathrm{RMS}}_{i,o,m}\left(j\right)\right)}^2},$$ (6)

where J_i,o,m is the number of points in the corresponding guidewire centerlines. We then determined the average RMS distance of the guidewire paths across users as

$${\mathrm{RMS}}_{i,o}=\frac{1}{M}\sum _{m=1}^M{\mathrm{RMS}}_{i,o,m}.$$ (6a)

The guidewire paths for each phantom i, material o, user m, and insertion n were subsequently averaged to achieve the mean guidewire path for each phantom and guidewire material and the RMS distance was determined for each guidewire centerline point and the complete centerline as:

$${\overline{g}}_{i,o}\left(j\right)=\frac{1}{M}\frac{1}{N}\sum _{m=1}^M\sum _{n=1}^N{g}_{i,o,m,n}\left(j\right),$$ (7)

$${\mathrm{RMS}}_{i,o}\left(j\right)=\frac{1}{M}\frac{1}{N}\sum _{m=1}^M\sum _{n=1}^N{(g_{i,o,m,n}\left(j\right)-{\overline{g}}_{i,o}\left(j\right))}^2,$$ (8)

$${\mathrm{RMS}}_{i,o}=\frac{1}{J}\sum _{j=1}^J{\left({\mathrm{RMS}}_{i,o}\left(j\right)\right)}^2.$$ (8a)

To determine the steel and plastic-coated guidewire paths similarity, the overlap t of the guidewire path distributions for the two materials was calculated for each phantom as:

$$t_i\left(j\right)=\frac{\mid {\overline{g}}_{i,pl}\left(j\right)-{\overline{g}}_{i,st}\left(j\right)\mid }{{\mathrm{RMS}}_{i,pl}\left(j\right)+{\mathrm{RMS}}_{i,st}\left(j\right)},$$ (9)

where a result of t_i(j)⩽1 indicates overlap between the two average guidewire paths distributions and t_i(j)>1 indicates little or no overlap exists between the two average guidewire paths at centerline point j. The percentage of the t_i(j) that lie below one is then taken as the percent overlap of the two guidewire paths.

RESULTS

Varying the region-growing thresholds of the phantoms did not result in large changes in alignment results (see Table 1). The rotation varied by less than 0.5° and the translation varied by less than 1.5 voxels and the overlap between the volumes was over 95% when the upper and lower thresholds were varied by ±4 voxel values. The overlap was determined as the ratio of the number of voxels resulting from a binary AND-ing of the two grown volumes after alignment and the number of grown voxels in one of the vessel phantoms. We believe these differences are due to partial volume effects. The same thresholds [45 (lower) and 60 (higher)] were used to grow the cylinder for each procedure. The rotations and translations between the grown cylinders were all less than 0.5° and less than one voxel, respectively, indicating the stability of the physical mounts and reconstruction of the phantoms.

Table 1.

Average intra- and inter-user variations of the guidewire paths for the different phantoms and guidewire materials.

Phantom	Material	Intra-user σi,o mm (voxel)	Inter-user σi,o mm (voxel)
A	Plastic coated	0.08 (0.3)	0.17 (0.6)
B	Plastic coated	0.09 (0.3)	0.19 (0.7)
C	Plastic coated	0.07 (0.25)	0.15 (0.5)
A	Steel	0.19 (0.7)	0.48 (1.7)
B	Steel	0.85 (3.0)	1.37 (4.9)
C	Steel	0.15 (0.5)	0.35 (1.25)

To assess guidewire path variability with insertion and user, we calculated the average intra-user variations [Eqs. 5, 6] and inter-user variations [Eqs. 7, 8] for all performed insertions by each user. The determined variations were averaged (for each guidewire material and phantom) [Eqs. 6a, 8a] (see Table 1). We see that the intra-user and inter-user variations for the plastic-coated guidewire paths are about a factor of 2 less than those for the steel, that the inter-user variations are about a factor of 2 more than the intra-user variations, and that the variations are about one −1.5 voxel or less, except for the steel guidewire in phantom B.

While the data in Table 1 are useful, more insight into the data may be obtained by considering the cumulative histograms of the square root of the variances calculated in Eqs. 5, 7 (Figs. 34, respectively) for the various phantoms and guidewire materials. In agreement with the results in Table 1, the cumulative histograms for the plastic-coated guidewires (Fig. 3) are similar for all phantoms. Notice that the intra- and inter-user cumulative histograms are similar as well (with the inter-user variations being slightly worse) and that 90% of variations are within 0.3 mm∕one voxel (about the diameter of the guidewire or 8% of the diameter of the vessel).

Figure 3.

(a) Cumulative histogram of intra-user variations in plastic-coated guidewire insertions for the three phantoms (A, B, and C) [Eq. 5]. (b) Cumulative histogram of inter-user variations in plastic-coated guidewire insertions for phantoms A, B, and C [Eq. 5].

Figure 4.

(a) Cumulative histogram of intra-user variations in steel guidewire paths for phantoms A, B and C [Eq. 8]. (b) Cumulative histograms of inter-user variations in steel guidewire insertions for phantoms A, B, and C [Eq. 8].

The cumulative histograms for the steel guidewires are different from those for the plastic-coated guidewires. The cumulative histograms show that about 90% of the variations are within 0.7 mm for phantoms A and C, but the cumulative histograms for phantom B are substantially different from the other phantoms for both intra- and inter-user variations. The increase in variations seen with the steel guidewire paths relative to the plastic-coated guidewire paths are probably due in part to the higher stiffness of the steel guidewire. The differences between the phantoms for the steel guidewire are related to its stiffness and the shape of the vessel.

We obtain similar results when we compare the average centerlines ${\overline{g}}_{i,o}\left(j\right)$ calculated as per Eq. 7. The RMS difference of the average steel and plastic-coated guidewire paths in each phantom is relatively low (3 voxels, 0.9 mm) for phantoms A and C, whereas it is over 6 voxels (1.9 mm) for phantom B (Table 2). These results all indicate a difference between steel and plastic-coated guidewire paths, albeit about 3 voxels on average. We find similar results when we consider the overlaps between the average steel and plastic-coated guidewire paths calculated as per Eq. 9; the paths in all three phantoms have less than 50% overlap and phantom B has the least overlap. Note the apparent inconsistences between RMS and overlap are due to the use of the variances in the denominator in Eq. 9. The variances calculated in Eq. 8 are similar to those shown in Table 1, which indicates that the variances for phantom B are substantially higher than those in phantoms A and C. If we define overlap as the paths being within a specific distance of each other (e.g., 0.5, 1, or 2 mm), we see that the percent overlap is greater than 50% for distances of 1 mm or more for phantoms A and C. Thus, the steel and plastic-coated guidewire paths appear to be similar, except for phantom B.

Table 2.

RMS distance and percentage overlap between average guidewire paths of steel and plastic coated determined in each phantom when using the variances [Eq. 8], 1 and 2 mm in the denominator in Eq. 9.

Phantom	A	B	C
RMS (mm)	0.87	1.90	0.68
% overlap (σ)	47	35	39
% overlap (0.5 mm)	35	3	37
% overlap (1 mm)	58	13	77
% overlap (2 mm)	100	48	98

To investigate the large variations occurring in phantom B for the steel guidewire, we compared the distance of the average guidewire path to the vessel wall to the variance at each index point j [Fig. 5a]. The distance to the vessel wall was determined by region growing. To facilitate comparison, both quantities were divided by the respective maximum values and converted to percentages. We see that the distance to the vessel wall correlates well with the variations for the various points along the guidewire path. Indeed, if we look at the data from the steel guidewires and all phantoms [Fig. 5b], we see that the variance appears to depend on the distance from the vessel wall and that for phantom B the distance from the vessel wall does become large (up to 2 mm, i.e., the guidewire crosses the center of the vessel).

Figure 5.

(a) Distance of ${\overline{g}}_{2,st}\left(j\right)$ to the vessel wall and the standard deviation σ_2,st(j) along the centerline for the steel guidewire in phantom B. (b) The average variation of the guidewire paths as a function of distance from the vessel wall for phantom B and phantoms A and C. As the guidewire path moves away from the vessel wall, the variation increases.

DISCUSSION

We have presented measurements of the reproducibility of guidewire paths in vessel phantoms. For plastic-coated guidewires, we found that the paths were reproducible to within approximately 0.1 mm (intra-user) and 0.2 mm (inter-user) for all phantoms. Note that we achieve variations smaller than the voxel size because of the precision of the centerline determination. We take these results to indicate that the plastic-coated guidewire path is reproducible and independent of insertion, user, and vessel itself. For steel guidewires, we found that the both intra- and inter-user variations in the paths were about a factor of 2 higher than those for plastic coated for two of the phantoms, but these variations were substantially higher for one of the phantoms, phantom B. For phantoms A and C, the paths of the steel guidewire were reproducible to within about 0.2 mm (intra-user) and 0.4 mm (inter-user) (note 0.4 mm=10% of the vessel diameter). In addition, for the phantoms A and C, the average guidewire paths of the plastic-coated and steel guidewires are comparable (RMS distance <0.9 mm and over 50% of the points lie within 1.0 mm of each other).

In exploring the data further, we found that the variations remained small for the plastic-coated guidewire even as the distance from the vessel wall increased, whereas the variations increased for the steel guidewire with the distance from the vessel wall. These trends were seen in all phantoms, with phantom B having the largest variations and largest distances from the vessel wall. The causes of these trends go beyond the scope of this paper; however, the largest variations and largest distances from the vessel wall do occur in a relatively long straight region near the midsection of phantom B. For this long, straight region, the constraints on the path are located at those points where the guidewire touches the vessel wall at either end of this region. It may be that the lack of constraints along this relatively large distance along the guidewire allows for greater variations.

In summary then, the paths of the two guidewires investigated here were reproducible to within about 0.5 mm independent of insertion and user and, average steel and plastic-coated guidewires were within 1 mm. The reproducibility does appear to depend somewhat on the vessel geometry itself, on the location of the guidewire in the vessel lumen, on the distance between constraining points (i.e., points where the guidewire touches the vessel wall), or on a combination of all three.

The conclusions derived from our research should be valid in the clinical in-vivo situation when the vessel is more rigid than the guidewire (either intrinsically or due to surrounding tissues). For those situations in which the guidewire-vessel interaction results in modification of the vessel path, it would seem that the guidewire path still might be reproducible. This interaction and the reproducibility of the guidewire path would be worth investigating but goes beyond the scope of the current work. To the best of our knowledge, no one has investigated the reproducibility issue before, and we see the reproducibility of the path as important, useful data for developing guidewire modeling techniques. These results can form the basis for developing methods to efficiently and accurately predict the guidewire position inside the vasculature prior to deformation using the finite element method¹⁰ or shortest path algorithms.³² Based on these paths, methods can be derived from existing finite element models facilitating guidewire stiffness and patient specific data (vascular geometry, vascular properties) to determine the vessel deformation resulting from the guidewire. Thus, we see this investigation as an important step in guidewire research.