1. Introduction

The human respiratory system consists of a complex network of hollow channels, extending from the nose and mouth, through the pharynx and vocal cords, and down to the terminal bronchioles. It may be afflicted with a range of pathologies, many affecting the size and shape of the airway lumen. For example, in the upper airway, obstructive sleep apnea manifests as repetitive collapse of the pharynx during sleep. In the lower airway, subglottic stenosis presents as a focal narrowing of the subglottic airway lumen. In both examples, accurate imaging of the airway lumen is critical for the guidance and assessment of treatment.

Anatomical optical coherence tomography (aOCT) [1–3] is an endoscopic optical imaging modality capable of producing quantitative measurements of airway lumen dimensions during interventions [4,5]. Unlike subsurface endoscopic OCT, which provides high-resolution images of tissues substructures over a range of 2-3 mm [6–10], aOCT provides quantification of airway lumen over a range of several centimeters. It has been demonstrated clinically in applications such as assessment of obstructive sleep apnea [11,12], lower airway stent selection [13,14] and assessment of in vivo airway elasticity [15].

During aOCT scanning, a rotating endoscopic probe is translated along the lumen and radial 2D B-scans are acquired and combined to reconstruct a 3D data volume. However, the airway can have a complex 3D structure, in which the lumen may taper, bend and bifurcate. A significant limitation of current endoscopic OCT imaging is that all imaging measurements are acquired relative to the location of the scanning probe, and do not account for changes in the 3D location and orientation of the scanning probe as it is translated through the airway. These endoscopic scans do not reflect the true 3D nature of the lumen, making it difficult to correlate endoscopic findings with other imaging modalities, such as X-ray CT. Augmenting these scans with the location and orientation of the scanning probe as it changes during the acquisition could significantly improve the clinician’s ability to interpret these measurements relative to the patient’s anatomy.

Magnetic tracking systems utilize electromagnetic induction to accurately calculate the position and orientation of a tracking sensor [16]. These systems consist of a transmitter that generates a pulsed electromagnetic field, and a sensor containing three perpendicular coils that carry the currents induced by the magnetic pulses. Since the electromagnetic field strength decreases with distance, the resulting change in induced current measured in each sensor coil allows the sensor’s relative position to be calculated. These systems are well-suited to endoscopic applications as they do not rely on line of sight, and have been demonstrated in conjunction with a range of other, non-optical imaging modalities [17–19].

In this paper, we propose a novel tracker-assisted aOCT probe, combining aOCT scanning with simultaneous magnetic position and orientation measurements, resulting in accurate anatomical 3D reconstructions. To the best of our knowledge, this paper presents the first endoscopic OCT experimental setup to incorporate a magnetic tracking system. We present an analysis of the accuracy of the probe’s location measurements using a proof-of-principle prototype, and demonstrate the probe’s potential with 3D scans of a phantom airway.

2. System design

A schematic diagram of the tracker-assisted aOCT probe is shown in Fig. 1 . The aOCT probe utilizes a 1.0 mm-diameter gradient-index lens to focus the light beam, and a right-angle prism of 0.7 mm width to redirect the beam perpendicular to the probe. These components are protected within a small metal housing (1.3 mm outer diameter), with an opening to permit the light beam to exit. The assembly is affixed to the end of a 1.8 m length of single-mode fiber encased in a biplex torque-transmission stainless steel coil to allow rotation. The entire assembly is housed in a transparent catheter with 2.2 mm outer diameter. The magnetic sensor was mounted with Parafilm (Pechiney Plastic, USA) on the catheter adjacent to the probe head. Both probe and catheter were translated simultaneously during scanning, keeping the relative distance between magnetic sensor and probe constant.

 

Fig. 1 Schematic of tracker-assisted aOCT probe.

Download Full Size | PDF

The aOCT scanner, described in detail elsewhere [1–3], utilizes a time-domain OCT system, with a frequency-domain optical delay line configured to achieve a scanning distance of 27 mm. A-scans are acquired at a frequency of 900 Hz. The light source has a center wavelength of 1304 nm with a bandwidth of 27.5 nm, giving a coherence length of 27.3 µm. The optimal transverse resolution was 102 µm at a distance of 5 mm from the probe head. To acquire data over a 3D volume of the airway, the probe is mechanically retracted with a linear pullback rate of 1.35 mm/s.

The magnetic tracking system used was the 3D Guidance medSAFE tracker (Ascension Technology, USA) with mid-range transmitter and 8 mm x 1.3 mm (length x outer diameter) tracking sensor. It uses pulsed DC magnetic fields [20], acquiring measurements at a rate of 240Hz.

3. System characterization

3.1 System accuracy assessment

Magnetic tracking systems are calibrated by comparing tracker results against a gold standard measurement system. The nominal absolute accuracy reported by the manufacturer for this system is 1.4 mm (position) and 0.5° (orientation) RMS, averaged over a volume of 20 cm < x < 58 cm, −30 cm < y < 30 cm, and −30 cm < z < 30 cm [21], where axes are defined relative to the orientation of the magnetic tracker transmitter. Note that tracker accuracy varies as a function of both position and orientation [22]. However, for reconstruction of endoscopic 3D OCT data, it is the relative accuracy that is critical, i.e., the position of the OCT probe relative to the start of the OCT acquisition.

We assessed the relative position accuracy against a motorized translation stage (LTA-HL Position Motorized Actuator, Newport, USA). Orientation accuracy was assessed with a set-up similar to the work of Day et al. [23], using a wooden frame system in order to minimize inhomogeneities in the magnetic field.

For position measurements, the tracked sensor was mounted via a wooden rod on a translation stage with a known starting position and orientation, orientated with the central axis of the probe directed towards the transmitter, and then translated and tracked over a range of 25 mm. This length was chosen as typical of aOCT pullback scans acquired in human airways. This was repeated along the x, y, and z axes of the magnetic tracker transmitter, and position measurements from both the tracker and the translation stage (gold standard) were compared. The process was repeated at three representative distances from the transmitter: 33 cm, 37 cm, and 41 cm. To assess the influence of the metal in the aOCT probe on tracker accuracy, the experiment was performed under three scenarios: using the bare sensor; with the sensor attached to the aOCT probe; and with the sensor attached to the aOCT probe and probe rotating. Results are presented in Table 1 . Rotation of the aOCT probe was not found to affect the accuracy results.

Table 1. RMS error in relative position of magnetic sensor for movement along the transmitter’s x-, y- and z-axes. RMS calculated over a pullback of 25mm, at different transmitter-sensor distances.

View Table | View all tables in this article

The most significant factor affecting accuracy was found to be distortion of the magnetic field by the metal housing that encased the aOCT probe. This distortion was minimized for measurements in the x direction due to the relative positions of the probe and magnetic sensor.

Orientation measurements were obtained at a single distance 30 cm from the magnetic transmitter along the transmitter’s x-axis. The tracker and aOCT probe were mounted on a wooden frame that could be set to a range of 12 predefined orientations ([0°, 360°] at intervals of 30°). The RMS error was calculated over the 12 measurements and the measurements repeated around each of the three axes of rotation: azimuth, elevation, and roll. The experiment was again performed under three scenarios: bare sensor; attached to stationary aOCT probe; and attached to rotating aOCT probe. Results are presented in Table 2 . Orientation accuracy was found to be the same in all three scenarios, with the probe having minimal effect on the measured angles. Additionally, our results showed comparable accuracy around all three axes of rotation.

Table 2. RMS error in relative orientation of magnetic sensor (bare sensor, stationary aOCT probe and rotating aOCT probe).

View Table | View all tables in this article

4. Phantom imaging

An airway phantom was constructed from non-metal pipes with diameters roughly comparable to the human trachea, as depicted in Fig. 2(a) . The upper (orange) pipe had an inner diameter of 19.9 mm, whilst the lower (white) pipe comprised sections with two different diameters: 25.0 mm near the interface and a narrower 20.5 mm further down the phantom. Between the two pipes, an artificial stenosis (focal narrowing) was introduced with modeling clay. The stenosis was ellipsoidal in shape with major and minor axis lengths of 16.4 mm and 6.6 mm, respectively. All diameters were validated prior to scanning using calipers.

 

Fig. 2 Airway phantom: (a) photograph with caliper measurements; and (b) tracker-assisted aOCT reconstruction showing true 3D orientation of the lumen.

Download Full Size | PDF

The tracker-assisted aOCT probe was used to image the phantom airway and reconstruct a 3D data volume showing the true 3D orientation of the lumen. Quantitative analysis of the lumen is presented in Table 3 , and illustrated in Fig. 2. The average error of diameter measurements obtained using the reconstruction was 1.1%, with a maximum error of 2.0%.

Table 3. Phantom dimension measurements using calipers and tracker-assisted aOCT.

View Table | View all tables in this article

To illustrate the significance of tracker-assisted reconstruction, Fig. 3 shows 3D visualizations of the reconstructed data set, both without (Fig. 3a) and with (Fig. 3b) the use of the magnetic tracker. In Fig. 3a, consecutive radial B-scans are simply stacked to form a cylindrical volume, taking no account of changes in orientation of the OCT probe. This naïve reconstruction is representative of all previously published endoscopic 3D OCT scans. In Fig. 3b, changes in the probe orientation and trajectory have been included in the reconstruction through the use of the magnetic tracker, showing the true 3D nature of the lumen. A fly-through animation of the tracker-assisted reconstruction is shown in Fig. 4 and available online, both in low resolution (Media 1) and high resolution (Media 2).

 

Fig. 3 aOCT reconstructions of the phantom: (a) without; and (b) with tracker assistance.

Download Full Size | PDF

 

Fig. 4 3D animation of the tracker-assisted aOCT reconstruction of the phantom. Low resolution (Media 1). High resolution (Media 2)

Download Full Size | PDF

5. Discussion

Results presented here demonstrate the potential of magnetic tracker-assisted reconstruction with endoscopic aOCT. This prototype system presents the first published endoscopic OCT set-up capable of reconstructing the true 3D shape of an airway. These reconstructions make possible the comparison of such scans with pre-operative imaging modalities such as X-ray CT, enabling assessment of changes in pathology (e.g., extent of stenosis) between pre- and intra-operative imaging. In addition, true 3D reconstruction of airway shape would facilitate longitudinal assessment of airway pathology by correlating endoscopic aOCT acquisitions.

The experimental set-up presented in this manuscript provides proof-of-principle results. Practical clinical implementation will require that the magnetic tracker sensor be fully enclosed within a protective catheter, and designed to fit within the working channel of a standard bronchoscope (typically 2.2 mm – 3.2 mm). The combined probe outer diameter could potentially be minimized by integrating the magnetic coils of the sensor with the probe’s focusing optics.

System accuracy has been assessed using an airway phantom with dimensions comparable to a human trachea. Notwithstanding the long scanning distances used in aOCT, the relative accuracy of sensor position was found to be sufficient for accurate reconstruction of airway lumens of these dimensions.

Our characterization of relative tracker accuracy in the presence of the aOCT metal probe highlights the importance of careful probe design. Whilst magnetic tracking systems are designed to be accurate in the presence of medical-grade stainless steel, the metallic housing for the aOCT probe was found to introduce a small degree of inaccuracy in position measurements. This was due to magnetic eddy currents induced in the metallic housing by each transmitter pulse. We note this had less impact in the x-dimension, compared to the y- and z-dimensions, due to relative placement of the probe. For some metals, such artifacts can be minimized by careful selection of the transmitter sampling rate. Many types of steel are ferromagnetic with high magnetic permeability, and so the induced magnetic eddy currents are sustained for a longer period of time than in materials with low magnetic permeability (e.g., aluminium). LaScalza et al. [24] showed that increasing the sensor sampling rate will lead to an improvement in accuracy of the tracking system in the presence of steel. Alternatively, careful selection of materials will greatly reduce such issues, considering both magnetic properties of the metal and safety requirements of the application. Austenitic stainless steel is non-magnetic and will have minimal impact on tracker accuracy.

The results presented here also quantified the impact on accuracy of distance between the magnetic transmitter and sensor. We have demonstrated that accuracy decreases as a function of increasing distance. This can be attributed to two factors: decrease in the magnetic field strength with the inverse square of the distance, resulting in a decreased signal-to-noise ratio; and a loss of overall magnetic field homogeneity at the edges of the recommended region of operation. Such effects can be reduced by minimizing the distance from the magnetic transmitter to the sensor. To facilitate this, some manufacturers offer flat transmitters, which may be placed immediately underneath the patient for use in the operating theatre.

The angular accuracy was observed to decrease as the sensor neared alignment with the z-axis of the transmitter. Two factors are responsible for this effect: the ambiguous (non-unique) representation of angles when the sensor is precisely aligned with the z-axis; and because small physical changes in magnetic sensor orientation at this alignment result in large changes to the angular measurements, especially the azimuth.

Finally, accuracy is limited by the discrete spatial resolution of the tracking system. For the medSAFE system, this is approximately 110µm. We note that by modeling the motion of the endoscopic probe, such as by assuming linear motion over a small temporal window, it may be possible to achieve higher accuracies through interpolation of the spatial position.

6. Conclusion

This paper has described a novel prototype tracker-assisted aOCT system for imaging hollow organs. This is the first published demonstration of a magnetically tracked endoscopic OCT probe, showing the true 3D lumen shape. The system operates by using a magnetic tracker to provide position and orientation data for reconstruction of aOCT radial B-scans. At 33 cm distance from the magnetic tracker’s transmitter, the optimal relative accuracy of the tracker-assisted probe was 60 µm, 174 µm and 215 µm in the x-, y- and z-dimensions, with axes defined relative to the transmitter. A full 3D reconstruction was presented of a lower airway phantom, showing the true 3D nature of the lumen. Phantom measurements of lumen diameter obtained using this reconstruction were within 2% accuracy of the gold standard values.