A navigation system for percutaneous needle interventions based on PET/CT images: Design, workflow and error analysis of soft tissue and bone punctures

Abstract

Percutaneous needle intervention based on PET/CT images is effective, but exposes the patient to unnecessary radiation due to the increased number of CT scans required. Computer assisted intervention can reduce the number of scans, but requires handling, matching and visualization of two different datasets. While one dataset is used for target definition according to metabolism, the other is used for instrument guidance according to anatomical structures. No navigation systems capable of handling such data and performing PET/CT image-based procedures while following clinically approved protocols for oncologic percutaneous interventions are available. The need for such systems is emphasized in scenarios where the target can be located in different types of tissue such as bone and soft tissue. These two tissues require different clinical protocols for puncturing and may therefore give rise to different problems during the navigated intervention.

Studies comparing the performance of navigated needle interventions targeting lesions located in these two types of tissue are not often found in the literature. Hence, this paper presents an optical navigation system for percutaneous needle interventions based on PET/CT images. The system provides viewers for guiding the physician to the target with real-time visualization of PET/CT datasets, and is able to handle targets located in both bone and soft tissue. The navigation system and the required clinical workflow were designed taking into consideration clinical protocols and requirements, and the system is thus operable by a single person, even during transition to the sterile phase. Both the system and the workflow were evaluated in an initial set of experiments simulating 41 lesions (23 located in bone tissue and 18 in soft tissue) in swine cadavers. We also measured and decomposed the overall system error into distinct error sources, which allowed for the identification of particularities involved in the process as well as highlighting the differences between bone and soft tissue punctures.

An overall average error of 4.23 mm and 3.07 mm for bone and soft tissue punctures, respectively, demonstrated the feasibility of using this system for such interventions. The proposed system workflow was shown to be effective in separating the preparation from the sterile phase, as well as in keeping the system manageable by a single operator. Among the distinct sources of error, the user error based on the system accuracy (defined as the distance from the planned target to the actual needle tip) appeared to be the most significant. Bone punctures showed higher user error, whereas soft tissue punctures showed higher tissue deformation error.

• PET/CT image guided intervention
• navigation
• computer aided intervention
• multimodal needle guidance
• bone
• soft tissue
• percutaneous needle intervention

Introduction

Percutaneous needle interventions (i.e., biopsy, wire marking, radiofrequency ablation, etc.) are widely used in medicine as minimally invasive alternatives to open surgical interventions. Such procedures are typically performed based on structural volumetric datasets such as, for example, those obtained using computed tomography (CT) and magnetic resonance imaging (MRI). However, medical imaging has changed fundamentally with the introduction of functional modalities such as Single Photon Emission Computed Tomography (SPECT) and Positron Emission Tomography (PET). By using combinations of structural and functional modalities (SPECT/CT and PET/CT), image understanding, and therefore diagnosis, can be optimized [1], [2]. With these techniques, even suspicious lesions without morphological correlation, i.e., those not evident in structural datasets, can be identified and verified histologically [3]. In multimodal image interventions, the functional image (e.g., PET or SPECT) provides information about the target, while the structural image (e.g., CT, MRI, etc.) provides anatomical information for geometric instrument guidance. Such procedures can be performed using a step-by-step technique (similar to CT-guided interventions). Currently, the intervention is planned based on a reference image comprising a single-bed PET/CT acquisition of the patient already positioned for needle intervention. This reference CT includes a reference grid placed on the patient's skin that is used to facilitate guidance of the needle (i.e., a cannula with an appropriate puncturing instrument) to the lesion. After identifying the grid and the lesion in the reference dataset, the needle is then oriented and introduced in small steps. Each step is alternated with control CT acquisitions to ascertain the needle's current position and orientation with respect to the grid. The control scans are performed until the target is reached. In the case of bone punctures, the trocar (i.e., the sharply pointed instrument that fits in the cannula and is used to puncture soft tissue) is replaced by a drill (i.e., the drilling instrument that fits in the cannula and is used to penetrate the bone) as soon as the needle reaches the bone surface, allowing the physician to drill to the target. The depth of the drill is controlled with the aid of distance holders (provided with the drilling kit) which help prevent drilling deeper than a specified value. Although effective, the step-by-step approach requires an excessive number of control CT scans and lacks a registration between the patient and image datasets that must therefore be performed mentally by the clinician. Consequently, the radiation burden to the patient is increased, along with the procedure time.

Stereotactic instrument guidance is a technique used to provide increased procedural accuracy in anatomical areas that cannot be viewed directly with the human eye. The method relies on the position-sensing of instruments and a known patient anatomy inside a navigated space in order to co-display virtual models of the instruments in the context of registered medical image datasets. The general principles of this technique are described in references [4] and [5], and have been applied to a variety of clinical areas. Today, navigation systems rely mainly on two different tracking technologies, optical and magnetic, to track or estimate the pose of a dynamic reference (DR) in space. A comparison between the two methodologies [6] shows that optical methods present higher accuracy but require a continuous line of sight between the tracked instruments and the camera. In contrast, electromagnetic tracking does not require a line of sight, but has lower accuracy and requires a wired connection to the tracked instruments. Moreover, the latter method is sensitive to ferromagnetic material present in the tracking environment and is therefore less robust than the optical method. In the area of needle guidance applications, many navigation systems have been developed to solve specific clinical problems, such as liver puncture [7], [8] and vertebroplasty procedures [9], as well as for general biopsy procedures [10], [11]. While some groups [9-11] rely on well established rigid approaches for registering the patient to the image and are therefore limited to targeting static anatomical regions, other groups have investigated non-rigid methods to gate breathing [7] or to model deformation [8] in order to target organs that can deform considerably during the procedure. A recent review by Cleary and Peters [12] shows that even though non-rigid methods are making progress for interventions in the thorax and abdomen, most current work is still based on the rigid approach.

Regardless of the method used, systems for interventions based on PET/CT images, which require manipulation and real-time visualization of multimodal datasets, are still not available. To address this problem, Militz et al. [13] demonstrated that needle guidance using multimodal datasets is feasible with standard commercial navigation systems, although manipulation of such images and operational flexibility is decreased. Other groups also reported multimodal approaches using combinations of pre- and intra-procedural image datasets for needle guidance [14], [15]. Phantom experiments have demonstrated the feasibility of using such navigation systems with multimodal images. However, the handling of multiple acquisitions of structural images as well as different sets of metabolic images while following clinically approved protocols for oncologic needle interventions, such as in reference [3], still requires investigation. Additionally, most of the presented systems do not distinguish clearly between lesions located in bone and those in soft tissue. Each puncture type follows a different clinical protocol and therefore cannot be treated identically by the navigation system, especially given the challenge of navigating the drill in the case of a bone lesion. Integration of a DR into the drill can be difficult, considering their shape and functionality. Consequently, when lesions are located in soft tissue, the needle can be guided all the way to the target, whereas for lesions located in bony tissue, needle guidance can only be used until the bone surface is encountered; thereafter, the orientation of the drill is known from the tracked cannula, but no information on the depth of the drill is provided. The distinct requirements of each puncture type might change the relevance of the error sources, as well as the outcome associated with each of type of lesion. Studies comparing results for navigation systems dealing with these two types of puncture are still needed.

To address these problems, we have developed an optical navigation system for percutaneous needle interventions based on PET/CT images which can handle multiple datasets as well as both types of punctures, and which follows clinically approved protocols. The system software provides viewers for guiding the physician to the target with real-time visualization of PET/CT datasets, and for estimation of the final drill position in the case of bone interventions. In addition, the system workflow is compatible with standard clinical protocols for needle interventions, and thus does not increase setup time or personnel requirements as the system is operable by a single person even during transition to the sterile phase. The preliminary version of this system, as presented in this paper, was designed to target static anatomical regions (those not affected by breathing, e.g., the pelvic region). Therefore, it could be evaluated in an initial set of pre-clinical experiments using simulated soft tissue and bone lesions in swine cadavers. Herein we present a detailed description of the system design, workflow and evaluation methods, followed by an analysis of the system accuracy based on an error decomposition approach, which enables us to investigate the different steps in the process and the particularities relating to soft tissue and bone punctures.

System description

A navigation system for percutaneous needle intervention based on PET/CT images and the workflow necessary for navigated interventions targeting bony and soft tissues were designed.

The navigation system consists of a computer, an optical position sensor (NDI Vicra camera, Northern Digital, Inc., Waterloo, Ontario), two touch screens (and/or a wireless mouse), a moveable cart, and a DR set. The later comprises a needle DR to be attached to the needle shaft (i.e., the back part of the cannula) to provide a tracking reference for the actual needle, a patient DR to be attached to the patient skin to provide a tracking reference for the patient, and a calibration unit used to calibrate the needle tip and shaft positions in relation to the needle DR. In general, the navigation system tracks the pose of each DR with the position sensor and co-displays a model of the needle along with the PET/CT image by applying the set of transformations described in Figure 1. This technique provides the physician with visual feedback for instrument guidance assistance. The passive optical tracking technology chosen relies on infrared retro-reflective spheres to estimate the pose of each tracked DR, and thus avoids the need for wired connections to the tracked instruments. A set of clamps provides quick connections between a wide variety of needle shafts and the needle DR. The DR set and the clamps were manufactured from medical-grade titanium and are therefore sterilizable.

Figure 1. Navigation system overview. The transformations ^camT_ndr and ^pdrT_cam are measured by the position sensor, whereas ^imgR_pdr is the patient-to-image registration.

The system software is divided into three modules: planning, calibration and registration, and instrument guidance. Each module is described below.

Planning

Three standard PET/CT planar viewers (coronal, sagittal and axial) provide the same functionality used in clinical routine, and one 3D viewer (Figure 2a) provides an overview of the procedure plan. Using the planar viewers, the user can set the target location precisely, and through the 3D viewer a needle entry point can be quickly chosen by selecting a position on the patient's skin. These two locations define the planned trajectory that can be verified on the trajectory viewer (Figure 2b). This viewer displays structures surrounding the planned path, which helps in the localization of possible anatomical obstacles and critical structures. The patient DR pose is defined in the image coordinate system by selecting the center of the retro-reflective spheres in the CT image. With the aid of the 3D viewer for coarse definition and the planar viewers for fine definition, this procedure takes less than 1.5 minutes. After planning, the user gains visual access to the tracked DR set, and a 3D viewer provides information about the camera viewing volume so that the optimal position of the camera can be found prior to the intervention.

Figure 2. (a) 3D viewer showing an overview of the planned situation and the orientation of the trajectory viewer. (b) Trajectory viewer showing the planned trajectory and possible obstacles. (c) Navigation platform beside the PET/CT scanner table. (d) Needle and calibration unit in calibration position.

Calibration and registration

The calibration unit is used together with the specific needle calibration adaptor for calibrating the needle tip location and axis orientation with respect to the needle DR. The calibration is performed by acquiring a set of poses of both the needle DR and the calibration unit, and then transforming the average calibration vector (defined by the calibration unit construction data in the CNC machine) to the needle DR (Figure 2d). To reduce and simplify the software interaction, the calibration process triggers automatically as soon as the calibration unit is shown to the position sensor in the calibration configuration (as in shown Figure 2d). The clamp used to attach the needle DR to the needle shaft defines a region of probable location of the needle. The system continuously verifies that this region matches the calibration vector of the calibration unit and triggers the start of the calibration procedure whenever such condition remains fulfilled for a couple of seconds. Thereafter, a set of poses for calibration is acquired. The calibration process is aborted when the user varies the position of the tip relative to the calibration vector during the pose acquisitions. Following needle calibration, the patient-to-image registration ^imgR_pdr is automatically calculated (as described in reference [16]) once the patient DR is visible. This transformation maps the position of the patient DR spheres detected by the position sensor to their pre-defined positions in the CT image, described above in the Planning module sub-section. The whole process takes approximately 10 seconds.

Instrument guidance

Three main viewers – the 3D viewer, the targeting viewer, and the needle plane viewer – provide visual information to help guide the needle to the target according to the planned trajectory. The 3D viewer (similar to that in Figure 2a, but also including the virtual needle) shows an overview of the intervention scenario which facilitates localization of the planned needle entry point. The targeting viewer (see the schematic in Figure 3a and an example in Figure 4a) helps the user align the needle with the planned trajectory and navigate to the target. In addition, the viewer allows the physician to assess the risk to structures around the planned trajectory. The viewer also shows the needle tip and shaft within a crosshair aligned to the planned trajectory, in addition to the remaining distance to the target. The needle alignment with the planned trajectory is achieved by bringing both tip and shaft to the center of the crosshair. The needle plane viewer (see the schematic in Figure 3b and an example in Figure 4b) helps the user foresee upcoming structures and estimate the final position of the needle. It shows the virtual needle and its projection to the target on the PET/CT plane that is vertically aligned with the needle and horizontally parallel to the x axis of the needle DR. The needle projection displayed by this viewer is especially important for bone interventions, in which physicians can only navigate the trocar needle as far as the bone surface, having no further information about the drill tip depth while penetrating the bone. By combining the projection and the required drilling depth (i.e., the distance from the cannula tip to the target), the physician can estimate where the drill will end up after drilling the remaining distance.

Figure 3. (a) Schematic of the targeting viewer. The virtual camera is aligned with the planned trajectory and looks towards the target. The needle tip and the needle shaft are projected back to the crosshair plane, and are respectively displayed as a cross and a circle that is complementary to the cross. Both are connected through a line. The viewer background displays the fused PET/CT slice orthogonal to the planned trajectory at the depth of the needle tip d, or alternatively 10 mm beyond the tip, d + 10. (b) Schematic of the needle plane viewer. The virtual camera looks towards the needle tip, and has its y axis parallel to the needle, and its x axis parallel to the cross-product between the needle and the z axis of the needle DR. The viewer background displays the PET/CT slice surrounding the needle axis. To aid the physician in predicting the final tip position if the needle continues in the current direction, a projection of the needle along the remaining distance d is displayed.

Figure 4. (a) The targeting viewer with the needle aligned with the planned trajectory and moving towards the target. The distance to the target is displayed in the left bar, and is also represented by the distance arrows approaching the center of the cross-hair target, which removes the necessity to deviate one's attention from the center of the viewer. (b) The needle plane viewer displaying the needle, its projection to the target, and the PET/CT slice along the needle axis.

The system is set up inside the intervention room beside the scanning table, as shown in Figure 2c, and is connected to the hospital network. A sterilized DR set is kept ready for use in case a biopsy is required.

The system workflow, like the standard non-navigated PET/CT intervention [3], begins after diagnosis and continues until the target is reached. The process is divided into two phases: non-sterile and sterile. The sterile phase commences at needle calibration (see Figure 5). The system software was designed such that there is no need for manual interaction with the mouse or touch screen during the sterile phase, allowing the whole procedure to be performed by a single person. The individual steps can be described as follows:

Figure 5. System workflow emphasizing the non-sterile and sterile phases and the particularities of bone and soft tissue interventions.

Non-sterile phase:

• Patient preparation. The patient is positioned, prepared and stabilized by means of a vacuum mattress on the scanner table, as presented in reference [17]. At this point, the lesion and the likely needle entry point positions are coarsely known by the physician, and the patient DR can therefore be placed near the intended target. This minimizes the effect of errors between the patient DR and the target such as the tracking error in the target (TRE), as described by Fitzpatrick et al. [18]. The physician is also advised to place the patient DR in regions not affected by respiration movement, which limits the system's use to specific areas, e.g., the pelvic region. The patient DR can be attached to the patient's skin by means of surgical tape, requiring no more time than the standard reference grid attachment. Lastly, the system cart is placed on the opposite side of the patient to the physician.

• Image acquisition. A small PET/CT image volume including the patient DR and the potential lesion is acquired. The dataset is assumed to be co-registered, as provided by most currently available PET/CT scanners (e.g., the Siemens Biograph 16 Hi-Rez).

• Image transfer. The image is exported from the scanner station to the navigation system through the network. No additional time beyond that required for the normal export of data to a planning station is required by this step.

• Planning. The planned trajectory, composed of the planned target and needle entry point, as well as the patient DR spheres, is chosen precisely using the viewers described previously in the Planning module sub-section. In the final steps before entering the sterile phase, the physician aligns the monitor with the planned trajectory and adjusts the camera to ensure that the patient DR and the needle will be visible throughout the intervention.

Sterile phase:

• Needle calibration. Based on the trajectory length, the appropriate needle length and strength are selected. The needle DR is clamped to the chosen needle's shaft, and the needle tip position and axis of orientation are calibrated relative to the needle DR.

• Patient to image registration. At this point, the patient is automatically registered to the image datasets in order to transform points from the patient anatomy to their corresponding positions in the image dataset.

• Instrument guidance. The physician then uses the visual feedback provided by the navigation system to steer the needle to the bone surface or target in the case of bone and soft tissue interventions, respectively.

• Bone surface. From this moment on, the trocar cannot continue because the target is placed inside the bone. Thus, the physician has to exchange the trocar for the drill.

• Check projection. Using the needle projection provided by the needle plane viewer, the physician re-orients the drill to make sure it will hit the target. Subsequently, the missing distance is checked to determine the required drill depth. The navigation system has no information about the tip depth, and the drill depth is thus controlled by the physician alone.

• Drill to target. Finally, the physician drills the missing distance towards the target.

• Target reached. The physician decides whether the target was sufficiently reached, and finishes the procedure (i.e., takes a sample, places the marking wire, etc.).

Experiments

To quantify error sources involved in the navigation process, a series of experiments involving bone and soft tissue punctures based on PET and CT modalities were performed in animal cadavers.

A navigated PET/CT pilot study using 4 swine cadavers (labeled S1, S2, S3 and S4) prepared with radioactive spots (cotton wool balls 1 mm in diameter with approximately 0.5 MBq ¹⁸F solution placed inside bone and soft tissue) was performed on site with consideration of the constraints faced during daily clinical routine (e.g., available space, sterility phase, and available procedural time). The first two swine were used for the simulation of 23 bone punctures, while the other two were used for the simulation of 18 soft tissue punctures. The lesions were distributed along the swine anatomy, excluding the legs and head.

Cadaver puncturing procedure

The animal was laid on the scanner table and a set of punctures were performed according to the workflow previously described in the System description section. To ensure repeatability of the target definition during the validation phase, the radioactive spot in the PET modality was adjusted to approximately 5 mm in diameter (using the window/level functionality) and the target was defined by visually fitting a circle around it using the three views (coronal, sagittal and axial). During the instrument guidance step, the physician was asked to decide the final position of the needle. After releasing the needle, a final image acquisition (PET/CT) with the needle in place was performed. In total, two datasets (each containing co-registered PET/CT) were obtained per puncture: a planning dataset used for planning the puncture and guiding the needle, and a validation dataset for checking the actual needle position. This process was repeated for every lesion.

Equipment

Image acquisition was accomplished with a Siemens Biograph 16 Hi-Rez with voxel size set to 1.37 × 1.37 × 1 mm and 2.67 × 2.67 × 2 mm (intrinsic spatial resolution of approximately 5 mm) for CT and PET, respectively. Both image modalities share the same coordinate system and were acquired right after one another. To eliminate or at least make insignificant the needle bending effect [19], a bone biopsy set with a coaxial needle of outer diameter 3.4 mm and length 130 mm with a distance holder of 5 mm, and a soft tissue biopsy set with a coaxial needle of outer diameter 2.1 mm (14 G) and length 145 mm were chosen for the study. The drill used for bone punctures was the same throughout the experiment.

Error decomposition

A common way of reporting the accuracy of navigation systems for needle guidance is to present the achieved overall error, e_ova, i.e., the distance from the target to the needle tip in the validation dataset. Unfortunately, this measure does not provide detailed information about the accuracy of the different system components involved in the process. To better understand the system behavior and indentify its particularities when applied to bone and soft tissue, we decomposed the overall error into individual sources. In this study, we divided the overall error into the following sources: e_pdr, i.e., the error introduced by a patient DR movement, e_tde, i.e., the internal tissue deformation error, and e_usy, i.e., the user-system error. The first source, e_pdr, appears when the patient DR moves relative to the target after the CT acquisition, which consequently causes a shift in the target as seen from the navigation system point of view. This error source has the same effect as the TRE. Nevertheless, it is conceptually different because TRE is defined as target shift caused by an inaccurate registration of a pair of points in a rigid structure containing the target. In this system, however, the points (i.e., the spheres in the patient DR) are not rigidly connected to the target. This may result in an e_pdr which is additional to the registration error. The second source, e_tde, is introduced by the interaction of the needle with the surrounding tissue [19], which might also cause a target shift undetected by the navigation system. The last source, e_usy, represents the user error based on the provided instrument and target tracking accuracy. The user error is determined by the physician's ability to define the patient DR spheres used for registration and to follow the guidance information provided by the system. The instrument and target tracking accuracy comprise the DR tracking accuracies achieved by the position sensor, the calibration accuracy and the TRE, as defined in reference [18], after registering the spheres defined by the user. The relation of the overall error vector to these individual source vectors is illustrated in Figure 6 and is formulated as follows:

Figure 6. Illustration of the target projection and error components. (a) Different projection methods: patient DR registration transformation R_pdr, rigid structure registration transformation R_rig, and machine coordinate system transformation R_mcs. (b) Locations of the target projected with different methods: patient DR registration transformation R_pdrP_plt, rigid structure registration transformation R_rigP_plt, and machine coordinate system transformation R_mcsP_plt. Additionally, the picture on the right illustrates how the different error vectors correlate to one another.

Measurement of individual error sources

In a second step, the datasets acquired during the punctures were post-processed to quantify components of the error involved in the process. The target was redefined in the validation dataset as P_vat, and the planned target, P_plt, was projected from the planning dataset to the validation dataset. A simple projection of P_plt using the common machine coordinate system transformation R_mcs is not sufficient to transfer the planned target to the validation dataset because it does not consider a possible movement of the animal between the planning acquisition and the validation acquisition to the validation acquisition (e.g., due to the physician-animal interaction while puncturing; see Figure 6 for an illustrative example). Hence, P_plt was projected using a transformation R_rig, based on a common reference within the body (assumed to be a rigid structure in the body, e.g., a vertebra [see Figure 6]). To project P_plt to the validation dataset as seen from the navigation system, a transformation R_pdr, based on the pose of the patient DR from both datasets, was used (see Figure 6). Finally, the actual needle tip, P_ntp, was defined in the validation dataset, and each error source was calculated as follows:whereThe transformation R_rig was calculated by registering a vertebra from the planning dataset to its corresponding position in the validation CT dataset. The transformation R_pdr was calculated by registering the patient DR from the planning dataset to its corresponding position within the validation CT dataset. The accuracy of the registrations was assessed by generating a surface of each structure (vertebra and patient DR) in both datasets, and by measuring the distance between them. The average and standard deviation distances between the surfaces of all registrations performed were below 0.1 mm for the vertebra and patient DR.

The measurement of individual error sources was performed using the planning dataset acquired prior to puncturing and the validation dataset acquired immediately after puncturing. This limited the evaluation to the scenario seen in the validation image, i.e., after releasing the puncturing needle. Therefore, the measured tissue deformation error, e_tde, captured only the permanent deformation caused in the tissue around the target, whereas the deformation present throughout the puncture would appear as part of the user-system error. In a similar line, the error due to patient DR motion, e_pdr, included only the permanent motion of the DR; any movement present throughout the puncture would appear as part of the user-system error. In addition, the user error could not be directly quantified by measuring the distance from the needle tip to the planned target according to the navigation system because for bone punctures the drill and its tip are not navigated. Finally, the TRE that represents the discrepancy between the patient-to-image registrations could not be isolated based only on the acquired datasets. Consequently, any target error resulting from a mis-registration between the patient and image was also captured as part of the user-system error.

Results

The navigation system based on multimodal images was developed and evaluated in a pilot study using swine cadavers. The navigation platform, which stands beside the PET/CT scanner, did not interfere with the standard image acquisition procedure or with the normal activity in the biopsy room. The data were successfully transferred from the scanner to the navigation platform through the network. The time required to transfer and load the images into the system was 4 min on average, which is equal to the time required to export the data to a planning station in the standard procedure. The preparation and planning of the intervention could be accomplished at the side of the patient and with complete separation from the sterile phase. The physician could operate the system without additional help. The actual instrument guidance time during puncture averaged 7.5 min and 5.5 min for targets located in bone and soft tissue, respectively. The bone punctures showed an increased time due to the drilling portion of the procedure. The overall error and the individual error components were also measured for each puncture of the different swine and are presented below.

The system presented an average overall accuracy of 3.7 ± 1.8 mm, which would be considered sufficient for the PET/CT scenarios. The average overall accuracy achieved for bone and soft tissue procedures was 4.2 ± 2.0 mm and 3.1 ± 1.2 mm, respectively. The best result for each puncture type is presented in Figure 7. Table I shows the average and standard deviation values for each measured error source. These results show that the system is applicable to both types of puncture, though its performance is better in soft tissue than in bone. The user-system error was the most prominent source for both types of puncture, with an average value of 3.9 ± 1.7 mm for bone punctures and 2.1 ± 0.7 mm for soft tissue. The error due to patient DR motion was more evident in the bone punctures, with an average value of 1.2 ± 1.0 mm, whereas the error due to tissue deformation was more evident in the soft tissue punctures, with an average value of 1.4 ± 0.9 mm.

Figure 7. (a) The best result for the bone punctures: puncture 14 in S1 with a planned trajectory length of 81 mm. (b) The best result for the soft tissue punctures: puncture 2 in S3 with a planned trajectory length of 98 mm.

Table I.  Average and standard deviation of the error sources (in mm).

	∥eova∥		∥eusy∥		∥epdr∥		∥etde∥
	Avg	Std	Avg	Std	Avg	Std	Avg	Std
S1	3.5	1.4	3.3	1.1	1.3	1.1	0.7	0.5
S2	5.3	2.3	4.8	2.2	1.1	0.7	1.0	0.6
S3	3.1	1.1	2.1	0.8	0.1	0.1	1.5	1.2
S4	3.0	1.4	2.1	0.7	0.4	0.5	1.2	0.7
Bone	4.2	2.0	3.9	1.7	1.2	1.0	0.8	0.6
Soft tissue	3.1	1.2	2.1	0.7	0.3	0.4	1.4	0.9
All	3.7	1.8	3.1	1.6	0.8	0.9	1.1	0.8

To facilitate a better understanding of how each individual error source contributes to the overall error, the correlation between the error sources and the overall error is presented graphically in Figure 8 and discussed below.

Figure 8. (a) and (b) show the different error sources per puncture for bone and soft tissue punctures, respectively, ordered by the overall error. (c) and (d) show the correlation between the error sources and the overall error for bone and soft tissue punctures, respectively. The lines represent the linear fitting of each error source.

Tissue deformation error analysis

The difference in tissue deformation behavior for both types of puncture is highlighted by Figure 8c and d. The trend line of this error shows a very weak correlation with the overall error for bone punctures, whereas for soft tissue this correlation is prominent. The effect of this error source, ∥e_tde∥, can be clearly seen in punctures 6 and 7 of S3 that were performed in soft tissue. This indicates an attempt to correct the needle direction when it is closer to the target, which causes tissue deformation and target motion due to the interaction with the needle. The tissue deformation was less evident for bone punctures, as a bone structure was harder to deform or dislocate. Nevertheless, this deformation is also evident in puncture 10 of S1 and puncture 1 of S2, which were performed in flexible bones, i.e., rib and the internal part of the tail, respectively. Hence, the interaction between needle and bone caused the whole bone to move.

Patient DR error analysis

Although the average error due to patient DR motion, ∥e_pdr∥, is higher for bone punctures, the correlation between this error source and the overall error is not evident for either puncture type (see Figure 8c and d). The trend line of this error shows a very weak correlation between these errors in both cases. Moreover, it shows a scattering of this error around the trend line of the bone punctures. The punctures in which this error was most evident were punctures 2 and 12 of S1. Further analysis of the CT image datasets showed that the patient DR location (abdomen) for these punctures was not optimal. The formation of gases expanded the abdomen between the time of planning the CT acquisition and the time of puncturing (including the validation CT acquisition), consequently moving the patient DR. This effect, together with the longer time needed for puncturing bone, increased the effect of ∥e_pdr∥ in this type of puncture. In addition, the necessary forces applied to the needle while puncturing and drilling in bone were higher, causing consequent motion of the surrounding tissue and the patient DR. Despite not being the most significant source of error, care should be taken to avoid the special cases faced here. The error analyses performed for this source indicate that the patient DR employed can be effective if well positioned (e.g., on regions unaffected by respiration).

User-system error analysis

The values obtained for ∥e_usy∥ show the ability of the physician to use the system with the provided tracking accuracy. The trend line of this error (see Figure 8c and d) shows a strong correlation with the overall error for both types of punctures; however, bone puncture results present a more evident correlation than those for soft tissue punctures. Two factors appear to have contributed to this difference: the lack of instrument guidance of the drill, and a possible sliding of the drill while penetrating the bone surface (as described by the physician during the experiment). The first factor is associated with the method the physician uses to ensure the correct depth of the drill in standard bone interventions. Five-millimeter distance holder rings are added to the drill to ensure a maximum depth, and the missing depth (<5 mm) is measured with the naked eye. The effect of not carefully verifying the drilling depth is highlighted by punctures 2, 3, 5 and 7 in S2. In these cases, the decomposition of ∥e_usy∥ values presented the greatest error component in the drilling direction (see Figure 9 for examples). The second factor is associated with the sharpness of the drill and with the direction of drilling, which is optimal at 90° to the bone surface. The depth of the target inside the bone appeared to have emphasized the sliding of the drill. Small changes in the needle orientation can cause a large deviation for targets located deep within the bone (see Figure 9c). However, these problems are expected to be reduced by providing the physician with smaller distance holder rings and sharper drills for penetrating the bone surface. A special case appeared in puncture 4 in S2: The planned trajectory of this trial passed through connective tissue or ligaments (whose presence was unknown to the physician) that caused the needle to deviate from the planned trajectory. Even though, the physician tried to correct the needle orientation, this unexpected tissue pushed it out of the planned path. This needle misplacement resulted in the greatest user-system error of our experiments, and demonstrated the importance of good trajectory planning. While the soft tissue punctures did not present user errors as high as those for the bone punctures, the results show that there is still room for improvement. One point that remains to be investigated is the learning curve of the physician using the system, which might reduce the user error and consequently the user-system error.

Figure 9. Results of punctures in S2 with high error in the drilling direction: (a) puncture 2; (b) puncture 3; (c) puncture 5; and (d) puncture 7.

The mean patient DR centroid-to-target distance for bone and soft tissue procedures was 137.3 ± 24.0 mm (range: 90.9 to 194.6 mm) and 193.9 ± 47.8 mm (range: 107.0 to 299.l6 mm), respectively. The mean planned trajectory length for bone and soft tissue was 65.6 ± 17.8 mm (range: 26.6 to 99.1 mm) and 94.0 ± 20.6 mm (range: 56.3 to 125.2 mm), respectively. The detailed correlation between these distances and each error source is presented graphically in Figure 10 and discussed below.

Figure 10. The different error sources per (a) patient DR to target distance for bone punctures; (b) patient DR to target distance for soft tissue punctures; (c) trajectory length of the bone punctures; and (d) trajectory length of the soft tissue punctures. The lines represent the linear fitting of each error source.

Error per patient DR to target distance

The large scattering of the points around the trend line of the overall error and the user-system error of the bone punctures (Figure 10a) makes it difficult to correlate them with this distance. This indicates that other factors are more influential regarding these errors than this distance. However, the error due to patient DR motion shows a considerable increase with such distance. For the soft tissue punctures (Figure 10b), a correlation can be seen for all error sources. The patient DR motion effect, as well as the user-system error, increases with increasing distance. The graphs show the estimated increase in error when the patient DR is placed sub-optimally away from the target. Unexpectedly, the soft tissue deformation showed an increase with this distance for both type of puncture. Nevertheless, no conclusion regarding this correlation could be made from the trials performed.

Error per trajectory length

The graphs in Figure 10c and d show no clear correlation between the length of the planned trajectory and most of the error sources for both puncture types. However, the user-system error shows a tendency to decrease with increasing target depth. Further data analysis showed that, coincidently during our experiments, targets with short trajectory length were mostly located in flexible bones (e.g., the ribs or close to the tail), while targets with long trajectory lengths were mostly located in stable bones (e.g., the vertebrae or pelvis). The fact that flexible bones are usually located closer to the skin is a possible explanation for the observed decrease in user-system error with increased path length. Any movement of flexible bones while drilling would cause the system to provide inaccurate depth information to the physician, thereby increasing the user error during drilling. Moreover, it would not appear as a tissue deformation error because the validation of this study is limited to the image acquisition performed after the puncture, and most of the flexible bones can partially recover their initial state after the forces applied for drilling are removed.

Discussion

We have presented an optical navigation system for percutaneous needle interventions based on multimodal images containing separate anatomical and target information such as PET/CT. The proposed system workflow integrating planning, calibration and registration, as well as instrument guidance as part of the intervention, was considered effective and proved to reduce the number of CT acquisitions required. In this study, only one image acquisition was required for instrument guidance. During the experiments, the software was shown to support the clinically applicable workflow by separating preparation and planning from sterile phases, thereby allowing the system to be managed by a single person. A comfortable arrangement between the system and the physician could be found for most of the punctures; however, to achieve such a configuration, the platform position had to be considered during the planning of the trajectory. The needle guidance was found to be more intuitive when the monitor with the visual information was aligned with the planned trajectory. The image data could be easily transferred from the scanning station to the navigation system through the network, requiring no additional work compared to the standard procedure. According to the physician, the planning module provided more reliability in the trajectory choice, as well as the possibility of defining paths out of the axial planes, which was not possible with the conventional approach. The structures surrounding the planned trajectory could be identified and considered during the planning process to ensure a safer path definition. The real-time multimodal visualization of the access path along with the region around the needle provided reliable spatial guidance, as well as enabling recognition of eminent obstacles (e.g., surrounding bones or important anatomical structures). The needle plane viewer was especially important for the bone punctures, helping to guide the drill with the projection of the needle. The automatic needle calibration supported the workflow by removing the need for physical interaction between the physician and the GUI during the sterile phase.

Finally, we have presented an analysis of the system accuracy based on a pilot study with swine cadavers. The results showed the feasibility of using the navigation system for interventions with targets located in regions unaffected by respiration. During the experiments, punctures in soft tissue performed slightly better than those in bone; nevertheless, both types yielded promising results. The error decomposition provided a better understanding of the source of each error and helped to identify points for system improvement. It also facilitated comparison between the two puncture types, pointing out the particularities of each of them, as well as supporting the prediction of expected error estimations in different scenarios. We believe that this comparison between needle guidance for targets within these two different types of tissue provides the research community with crucial considerations for the development of a navigation system for use in such interventions. Examples of such considerations include the difficulty of penetrating the bone surface, which increases the final user error and consequently the user-system error, and the significance of the individual error sources for each different puncture type.

Our preliminary results motivated us to further develop the presented system and to commence clinical trials. In addition to improving the points identified in this study, work has been done to modify the rigid patient DR to a set of single markers in order to incorporate deformation detection into the system.

Acknowledgments

The authors would like to thank Kate A. Gavaghan for proofreading the paper.

Declaration of interest: The authors report no conflicts of interest.