OCT images have been obtained by using the fabricated OCT catheter based on a piezotube-actuated weighted fiber cantilever. To minimize the environmental effect and nonlinear phenomena encountered with the weighted fiber cantilever, we have developed a 2D scan strategy different from those used with the fiber cantilever in previous research. Our method of driving the device uses a single-frequency sine signal for each driving axis, which is sufficiently far from the peak resonance frequency. This semi-resonant Lissajous scan provides us with more robust scan characteristics demanded for the practical endoscope applications.
At first, 1D-scan OCT imaging was demonstrated with our fabricated probe to test the basic capability of 2D OCT imaging. In this study, a spectral-domain OCT (SD-OCT) system based on a dispersive grating and a high-speed InGaAs line camera (1,024 pixels) was used to acquire OCT images at the 1.3-μm band. The axial resolution of the OCT system was 11.2 μm in air.
shows the raw OCT image (a) and the horizontally re-sampled image (b) of a human finger tip. All the OCT images in this report were displayed on a white background (inverted). The imaged area was 2.1 mm (vertical) × 2.2 mm (horizontal) in full. Only the X driving axis of the scanning catheter was used to acquire this image. The r.m.s. voltage of the applied sine signal was 70 Vrms at a semi-resonance frequency of 62.76 Hz. The A-line rate was 47,000 lines per second. So, a 2D image contained 374 A-lines in the raw data. Because the scan was performed by a sine function of time, the B-scan needed to be re-sampled to get a correct scale along the horizontal direction. A simple linear interpolation was used to calculate the re-sampled A-line data with the timing information of the sinusoidal B-scan. To achieve this, we needed a synchronization signal that contained the timing information of the scan’s start time. Instead of using a timing signal taken from the electric driving signal, we extracted the timing information from the acquired OCT image data directly. As the scan was given as a sinusoidal function, similar A-lines were repeated at the peak of the scan. The A-lines at both edges in were blurred horizontally for this reason. The first A-line in a 2D image could be determined by an A-line that had the maximum correlation with the adjacent A-lines.
Raw OCT image (a) and the re-sampled image (b) of a human finger tip, given in a white-background display mode. The darkness of a pixel corresponds to the reflectivity scaled in dB.
A 2D scan of the fiber cantilever was obtained by a low-order Lissajous scan in which each axis was driven at a slightly different frequency without any driving signal modulation.
shows a schematic Lissajous scan pattern (solid blue lines) in the XY plane laid over an OCT en face image. In this schematic, the Lissajous pattern is illustrated as a sum of four ellipses with step-wise phase increases. A line and a circle are regarded as special types of ellipses in this description. The actual scan pattern of the 2D scan has a continuously increasing phase with an even larger number of open ellipses in series. We may approximate the full curve as a sum of ellipses for simplicity. The frequency difference between the two driving axes determines the shape of the pattern and the temporal period by which the pattern evolves and repeats. The full period of the Lissajous pattern evolution is given by TL = 1/|fx−fy| where fx is the operation frequency of the X-scan axis and fy is that of the Y-scan axis. The scan pattern evolves from a line to a circle during the first quarter cycle (TL/4) as shown in the left-hand side of . As seen in the right, it evolves again from the circle to a line perpendicular to the beginning line on the left during the second quarter cycle. And this evolution repeats again in an inverse manner during the following half cycle (TL/2), returning back to the start point denoted by S. A half cycle completes a 2D scan for the full square area. Previous research that utilized the Lissajous scan strategy conventionally used the half cycle as a 2D scan cycle to acquire the full square area. However, it is more economical to take just a quarter cycle that completes a circular scan area, abandoning the two corners of the square that are not scanned in the following quarter cycle. In this research, a quarter Lissajous cycle is regarded as a 2D scan cycle for which a 2D cantilever scan is completed to acquire a 3D OCT tomogram.
Schematic Lissajous scan pattern (blue lines) in the XY plane laid over an OCT en face image: the first quarter cycle (left) and the second quarter cycle (right).
By modifying the conventional OCT terminology for our case, a 2D scan in the XY plane is said to consist of multiple elliptic B-scans or B-scan ellipses. Each B-scan has the shape of an ellipse or, more accurately, an ellipse-like open curve running by 360 degrees. So, the running B-scan ellipse resembles two scan lines of the raster 2D scan strategy except that it is curved. Especially for the central area of visual importance, our Lissajous scan almost looks like raster scan lines running together in opposite directions. This raster-like feature may give a higher tolerance against a sample motion to our endoscope catheter. In this report, the linear B-scan at the first or the last in a quarter Lissajous cycle is called the primary B-scan of each 2D imaging cycle. As shown in , we define X’ axis as an axis of the primary B-scan, which is tilted by ~45 degrees with respect to the actuator’s X-scan axis. The Y’ axis is defined by the axis perpendicular to the X’ axis.
For the semi-resonant Lissajous scan, the X axis was driven at 62.76 Hz while the Y axis was driven at 62.60 Hz with a driving amplitude of 70 Vrms. The Lissajous period (TL) was 6.25 s and, hence, it took 1.56 s (TL/4) to acquire a 3D tomogram for a quarter Lissajous cycle. The A-line rate was 33.7 kHz, which supports 2 × 269 A-lines per B-scan ellipse in full.
shows the raw 2D OCT image of the primary B-scan (a) and that of a B-scan when it was almost circular (b). The sample was a human finger tip. The information of scan timing was extracted from the OCT image data in a similar way as explained above. As clearly seen in , a 2D OCT image of a primary B-scan has inversion (mirror) symmetry because of its linear scan pattern. Thus, the primary B-scan is characterized by nearly the same 2D OCT image found in the other half with the opposite horizontal direction. The axis of this symmetry corresponds to a scan start point denoted by S or S’ in the 2D scan pattern of . This start A-line was searched by the maximum image correlation of the two partial images of the primary B-scan divided by the start A-line in which one of them was inverted horizontally. For simplicity, only 20 horizontal lines, sampled sparsely in A-lines, were used to measure the image correlation in this searching process for a start point. After determining the starting A-line of the first frame, the raw data were re-sampled according to the Lissajous pattern by using a simple linear interpolation.
Raw 2D OCT image of the primary B-scan (a) and that of a B-scan ellipse when it was almost circular (b). The sample is a human finger tip.
A 3D tomogram was successfully reconstructed with a timing signal extracted by the image correlations.
shows the reconstructed OCT en face images [(a) to (f)], and the 3D-rendered tomograms with different view angles [(g) and (h)]. The imaged volume was a cylinder of diameter 2.2 mm and axial depth 2.1 mm acquired in 1.56 seconds. The image quality was quite comparable to that of the conventional OCT system that uses bulky galvo-mirror scanners. No significant distortion of the sample’s morphology was observed. The channeled morphology of the finger tip surface and the inner dermal layers was imaged as parallel stripes in the en face views. Note that no calibration of the scan pattern was applied to the image reconstruction. In order to confirm the scan pattern of no deformation, a well-defined straight edge was imaged by our catheter. An IR detection card cut on a side was used for this experiment to make sure the imaged area contained the straight edge. shows the rendered 3D tomogram of the IR detection card on the edge. The linear edge of the top plastic cover verified that our system did not exhibit any significant morphological deformation.
Reconstructed OCT en face images [(a) to (f)], and the 3D-rendered tomograms with different view angles [(g) and (h)]. A rendered 3D tomogram of an IR detection card with a straightly cut edge (i) is shown together.
It has been found that the raster-like feature of the Lissajous scan makes the image less vulnerable to the sample’s bulk motion.
shows an OCT en face image before cutting the corner areas out (a) and the circular scan area laid over the same image (b). The blue dotted circle in depicts the scan circle area which is denoted by (1). The two corner areas denoted by (2) and (3) were erased in en face images of . At the boundary of the circular scan area and the corner area (arcs between (1) and (2), or (1) and (3) in ), the morphology is not continued but rather abruptly cut by the boundary. This is explained by the bulk motion of the sample that occurred during the data acquisition time. As shown in , the B-scan ellipses do not meet each other inside the circular scan area but they intersect in the corner area. This means a point had been over-sampled there or scanned two times in a scan cycle with a relatively large temporal interval. So if the sample moved during the interval, the A-line data for a scan point out of the circle would be influenced strongly by the motion and would appear to have a mean value of two A-lines acquired at different moments. This effect of the sample motion must be the most significant at the diagonal area where the temporal interval between the two scans is the largest. As seen in , a point located inside the circle on the diagonal line had been sampled first at the beginning stage of the Lissajous scan. But a neighboring point located just out of the circle contained data acquired in the final stage. The maximum temporal interval between the neighboring areas could be as long as the 2D scan cycle, 1.56 s in our case. This interval is long enough to be influenced by the sample motion. Note that a transverse motion of 0.1 mm per second can cause this level of discontinuity, which seems inevitable for the endoscopic OCT application in practice. This observation gives another good reason for the Lissajous scan to take the circular scan area. The over-sampling problem has been avoided effectively by the circular Lissajous scan. Especially, the central area in the XY plane, normally regarded as the most important part, was scanned only once (except for the primary B-scan) with a nearly uniform sampling distance in a raster-like scan manner. There, the temporal interval between the neighboring areas was kept as small as possible, owing to the relatively low density of sampled A-lines at the center. This feature sufficiently explains the good image quality obtained in .
OCT en face image before cutting the corner areas out (a) and the scan circle area laid over the same image (b). The blue dotted circle depicts the circular scan area (1) and two corner areas [(2) and (3)].
The vulnerability of the corner area to the motion artifact can be confirmed by the excessively high density of the sampling points. By the parameters used in our 3D OCT imaging, the distribution of the sampling point density was calculated for the XY scan plane by geometric considerations. The scan area was divided into 100 × 100 μm2 cells and the number of A-lines dedicated to each cell was counted for the Lissajous scan pattern.
shows the sampling point density in color map (a) and its cut view along a horizontal line that passes the center. It was found that the central part of the circular scan area exhibited a depressed but nearly uniform sampling point density (~40 A-lines in average and 38 A-lines at minimum, per 100 × 100 μm2 for the central part), while the corner areas had an excessively high density that could result in a higher vulnerability to the motion artifact. It was also found that half of the scan time is abandoned, which was dedicated to the two corner areas.
Sampling point density in color map (a) and its cut view along a horizontal line that passes the center.