Handheld optical palpation of turbid tissue with motion-artifact correction

Handheld imaging probes are needed to extend the clinical translation of optical elastography to in vivo applications, yet such probes have received little attention. In this paper, we present the first demonstration of optical palpation using a handheld probe. Optical palpation is a variant of optical elastography that uses three-dimensional optical coherence tomography (3D-OCT) to provide maps of stress at the tissue surface under static compression. Using this technique, stiff features present beneath the surface of turbid tissues are identified, providing mechanical contrast complementary to the optical contrast provided by OCT. However, during handheld operation, relative motion between the probe and the tissue can induce motion artifact, causing spatial distortion of 3D-OCT and in turn, optical palpation images. We overcome this issue using a novel, dual-function bi-layer that provides both a fiducial marker for co-registration and a compliant section for estimation of the stress at the tissue surface. Co-registration of digital photographs of the bi-layer laid out over the tissue surface is used to measure and correct for motion in the lateral (xy) plane. We also demonstrate, for the first time, that optical palpation can be used as a method for monitoring pressure applied to the tissue during handheld operation, thus providing a more repeatable and robust imaging technique between different users. Handheld optical palpation is demonstrated on a structured phantom, in vivo human skin and excised human breast tissue. In each case, image quality comparable to bench-top 3D-OCT and optical palpation is achieved. © 2018 Optical Society of America under the terms of the OSA Open Access Publishing Agreement

The ability to perform optical elastography with a handheld probe is needed for extension to clinical use [1], yet such imaging is challenging, largely because of motion artifact distorting the image and causing a time-dependent variation in mechanical loading. There have been a number of in vivo demonstrations of optical elastography [8][9][10][11], however these have largely been performed by benchtop systems or semi-mounted probes, artificially removing motion artifact. Transition of optical elastography to handheld probes operated freely by the user is key to enabling wide application to in vivo and clinical scenarios, as is the case in ultrasound elastography [12][13][14]. Recently, a handheld optical coherence elastography (OCE) probe using manual compression has been demonstrated on excised tissue [15]. In this approach, motion artifact along the depth axis is accounted for by utilizing a noise-tolerant vector-method [16] for strain estimation and calculating an inter-frame cumulative strain. As this approach used phase-sensitive detection, it is very sensitive to motion artifact. To overcome this, in [15], hundreds of B-scans were averaged to generate elastograms, making it challenging to implement clinically. In addition, as this technique is not readily extendable to 3D, and therefore, en face imaging, it is limited in its ability to survey large tissue areas in clinically relevant timeframes.
In this paper, we demonstrate, for the first time, optical palpation, a variant of optical elastography, using a handheld probe unsupported by any apparatus. Optical palpation [17] is a tactile imaging technique that generates maps, referred to as optical palpograms, of the axial stress at the surface of tissue under static compression. The stress is calculated by measuring the deformation of a translucent, compliant layer with known mechanical properties using three-dimensional optical coherence tomography (3D-OCT). Elevated regions of stress indicate stiffer underlying features, thus if compression is kept approximately constant between samples, stress may be used to distinguish tissue types. Particularly in turbid tissues, where optical contrast alone is often insufficient to distinguish features of interest [18], optical palpation has been demonstrated to enhance visualization by providing complementary mechanical contrast [19,9]. Importantly, as optical palpation does not rely on dense oversampling, as is the case with phase-sensitive approaches [15,20], acquisition times can be as rapid as in standard 3D-OCT. In addition, as optical palpation provides en face images, it enables large surface areas (up to 16 × 16 mm [19]) to be scanned in one acquisition.
Key to realizing freehand optical palpation is overcoming motion artifact, caused by movement of the user's hand during scanning, or in the case of in vivo imaging, involuntary movement of the patient. Relative motion between the probe and tissue typically causes spatial distortion of 3D-OCT volumes which, in turn, distorts optical palpograms. This can degrade the appearance and size of tissue features, for example, sub-surface blood vessels and areas of tumor, and needs to be corrected to ensure both faithful reconstruction of tissue structure, and proper estimation of surface stress. In optical elastography to date, motionartifact has only been accounted for along the depth axis, and motion correction of handheld optical palpograms has not yet been demonstrated. There has, however, been several demonstrations of motion correction in 3D-OCT [21][22][23][24][25], based on information either encoded in the OCT images themselves, or provided by some concurrent tool. One such study used the registration of multiple scanning laser ophthalmoscope images acquired during the handheld 3D-OCT scans to measure and correct for motion in the lateral (xy) plane [26]. This registration, however, relied on the detection and alignment of vessels, and as such, this result may not be extendable to imaging of more turbid tissue, in which such key features may not be present. Another method estimates and corrects for motion using a fiducial marker, a rigid object of known shape and size laid out over the tissue of interest, and image correlation of consecutively acquired B-scans [27]. Although this removes the need for key features on the tissue surface, the rigid metal fiducial marker would obscure the mechanical contrast of the tissue, making it unsuitable for extension to optical palpation.
To demonstrate handheld optical palpation, we overcome these pivotal issues by implementing a novel, dual-function bi-layer, simultaneously performing both stress measurement and motion correction. Placed on the tissue surface, a compliant layer of soft material can be deformed in order to perform optical palpation, while a fiducial marker is embedded in a secondary top layer. This top layer is made of stiffer silicone to ensure that it is not deformed by hand motion. In our approach, motion correction is performed using a digital camera in-built in the handheld OCT probe to track motion in the lateral plane, relative to the tissue surface. Relative offsets and rotations between successive OCT B-scans are measured by simultaneously acquiring a photograph of the tissue surface for each B-scan in the OCT volume. We then run a pixel-intensity correlation algorithm on successive pairs of photographs, facilitating lateral motion estimation to within ∼16 μm. The corresponding Bscan is then transformed accordingly, for each B-scan in the OCT volume, forming a more accurate representation of the tissue geometry. This motion-corrected 3D-OCT volume enables us to generate motion-corrected optical palpograms, providing mechanical contrast complementary to the optical contrast provided by OCT.
We also show, for the first time, that optical palpation can be used to monitor pressure applied to the tissue surface during handheld scanning. This is an important development for handheld OCT in general, as it has been reported that pressure can affect the rate of blood flow and can warp structures imaged by OCT [28,29]. In addition, our technique can be used to ensure that applied pressure is below the pain threshold of human subjects (0.4 MPa [30]), to avoid discomfort [10]. We achieve this by computing and comparing the average stress in the area of interest, enabling more repeatable and consistent OCT imaging with handheld probes across different patients and operators.
To demonstrate handheld optical palpation, we present results from a structured phantom, in vivo scans of human skin, and excised human breast tissue. In each case, we show substantial reduction in motion artifact in 3D-OCT and hence optical palpation, providing accurate contrast of key tissue features. We also demonstrate optical palpation as a method for monitoring pressure applied to the tissue by the user, providing insight into the ability to compare scans. We believe that this work is an important development towards routine optical elastography with handheld probes, expediting development towards in vivo clinical application.

The handheld OCT probe
Scanning was performed using a commercial, spectral-domain OCT system (TELESTO III, Thorlabs Inc., Newton, NJ, USA) operating at a center wavelength of 1300 nm. The full width at half maximum (FWHM) axial and lateral resolutions were measured to be 5.5 μm (in free space) and 14.4 μm, respectively. The objective scanning lens (LK30, Thorlabs) has a working distance of 22 mm and the acquired imaging volume was 7 × 7 × 2.5 mm in x, y and z, respectively. The sample arm comprised a handheld OCT probe (OCTH-1300, Thorlabs). Figure 1(a) shows a photograph of the probe, with overlaid axes which can be used to describe motion relative to the tissue; six degrees of freedom, three axes of translation (x, y, z), corresponding to the fast (x), slow (y) and depth (z) scan directions, respectively, and three rotational directions (pitch, yaw, roll) which pivot around these axes. A piezoelectric ring actuator (PA), similar to that previously used for optical coherence elastography (OCE) [21], is fixed to the probe using a threaded connection, giving the probe a total weight of ∼0.3 kg. Although OCE was not performed in this study, this setup is designed to enable extension to such imaging in the future. A VIS-NIR AR coated glass imaging window (Edmund Optics) of diameter 20 mm and thickness 2 mm was attached to the ring actuator using wax. To perform scanning, contact is made between the window and the bi-layer, as described in Sec. 2.3, which is placed on the tissue surface. The reference mirror was removed from the handheld probe to facil the glass ima described prev A schema probe contain same optical the 7 mm × area of the rin second, with a

Photogra
Acquiring ph quantify shif (see Fig. 1(a form of a mat of the same f frame), i.e., co These 807 registration al contact and is feature that as Unlike in typically mak applied for Independent t conducted us monitor probe effect on the x the OCT syste y-axis within normal to the optical palpat seen to have significant, ca images). As s considers mot

Correctin
In order to co calculated usi maps each en versa, was det We demo structured ph surface is sho its location in location to its map this bac   Figure 5(a) sh Fig. 5(b) show . The rigid mo ng all B-scans Fig. 4 Figure 6( uncorrected e inclusion cann optical palpog outside the fi pth, gives the nstruction, as i scanning exact comparison wit re shape of the . 6(f), caused b created an ar the motion cor d during this sc he x direction on in the y dire e scan. A lack d. The dark ar ] effects from at the edges of operator's han w in Fig. 6 Fig. 4(c). E bottom edge of sed an area of the accuracy n in Fig. 6(f) hromatic doubl ted OCT imag ntarily towards e central regio Therefore, mo ensity (D), even nd motion in (a) ographs captured b phantom surface -artifact-free refer ted, and (f) motion palpograms. The he average stress interpolation inac optical palpo in Fig. 6(e), e motion-corre wn in Fig. 6 ly to the pe of the nerate the ), the area ntified by higher stress (∼7 kPa) than the surrounding phantom material (∼3 kPa), and its geometry is now consistent with the motion-corrected en face OCT image shown in Fig. 6(f). The average stress (or pressure), σ , applied by the user during this scan is 3.78 kPa, calculated by averaging all stress values in the cropped optical palpogram. This value is useful in quantitatively guiding operators with applying adequate pre-load compression and assessing the ability to compare with other handheld scans of this phantom. Since optical palpation only relies on 3D-OCT to measure the deformation of the transparent compliant layer, it is less susceptible to error caused by small, localized OCT SNR variations. As such, the image artifacts (A,D), highlighted in Fig. 6(f), are not noticeable in Fig. 6(h). Figure 7 shows an in vivo handheld scan of the fingertip of a 21-year-old Caucasian female. The silicone layer was placed on the finger, with care taken to minimize slip during scanning. Figure 7(a) shows an uncorrected en face OCT image at a depth of ∼0.4 mm below the skin surface. The fingerprint pattern can be identified by the stripe-like texture and the fiducial marker by its low OCT intensity, however, the shape of both is clearly distorted by hand motion during the scan. Transforming this image using our motion correction technique generates the image shown in Fig. 7(b). The circular shape of the fiducial marker is accurately reconstructed, and the lines of the fingerprint are now more easily recognizable. The small disruption in this pattern (A) is likely caused by the fiducial marker moving slightly with respect to the skin during the scan. This distorts the correction of the underlying tissue as the photograph co-registration preferentially aligns the fiducial marker in each photograph, due to its high contrast. Figure 7(c) shows the uncorrected optical palpogram, and demonstrates similar distortion to the en face OCT image (Fig. 7(a)). The area outside of the fiducial marker has been masked in grey. Processing the motion-corrected 3D-OCT volume for optical palpation provides the optical palpogram shown in Fig. 7(d). This result can be used to see that the operator applied less pressure to the fingertip than the structured phantom (shown in Fig. 6), an average of 2.91 kPa. The fingerprint pattern can be identified in the image by the stress contrast caused mainly by the ridges and troughs of the finger pad, i.e., surface topology. Figure 7(e) shows a photograph of the fingertip with the fiducial marker in place. The inset highlights the correspondence of both the corrected OCT and optical palpograms to the area scanned. Additionally, this photograph shows a small horizontal crease line (C) present on the fingertip; not readily visible in the OCT image but revealed by optical palpation. Scattered small vertical lines of high stress (E) are also present, indicating areas where the automatic algorithm for compliant layer/tissue edge detection, used in the calculation of stress (see Section 2.1), failed due to optical artifacts near the layer-tissue interface.

Human b
A sample of f carried out at the patient, an Heath Servic Following su pathology lab over the tissu in Fig. 8(a). surgery, and surgical scar. of the fiduci specimen take the fiducial m returned for adipose (P), t view of the b images with th Figure 8(d the tissue sur difficulty in c shape has bee less spatially clear that the scan of the str 7. Human fingertip b) corresponding n corrected optica n grey). σ is the g the scan.  Fig. 8(b) to ssue micro-arch (A) caused by i below skin surface (d) corresponding r has been masked plied to the tissue n place. The inset n the skin on the mastectomy p nsent was obtai the South Met mittee (FSH-20 g theater to th icone layer wa mm × 8 mm, location of a nding to a we icated by the p eld OCT imag 5], with the po imen was imm in Fig. 8(c)

Discussion
The first demonstration of handheld optical palpation is presented in this paper. A main challenge we overcame was correction of operator-and tissue-induced motion artifact occurring during the scan. Importantly, in this study, the operator used a contact probe held freely in their hand. This is in contrast to prior demonstrations of optical palpation which used a support apparatus, such as an articulating mechanical arm [9] or mechanical stages [10] which are cumbersome for use in clinical applications.
Unlike for x, y, and roll motion, independent testing of z, pitch and yaw motion, while scanning with the OCT probe in contact with turbid tissue, revealed negligible motion-artifact on the 3D-OCT signal intensity. For pitch and yaw, this was confirmed in the structured phantom scan, by the consistent en face location of the inclusion edges throughout motioncorrected OCT volume. Comparisons of y-z profiles of the motion-corrected and mounted phantom scan, also confirmed negligible z-shifts. The handheld OCT probe used for this testing was fitted with an actuator to enable future extension to OCE [20] and quantitative micro-elastography [31] which rely on the fidelity of 3D-OCT scans. Since these techniques rely on phase-sensitive detection, they are significantly more sensitive to motion in z, as in the recent paper by Zaitsev et al. [15], than optical palpation. In ongoing work, we will assess existing and novel methods to also account for z-motion in OCE, and the possibility of extending our motion correction technique to image strain and elasticity using a handheld OCT probe.
Unlike in many previously demonstrated motion correction approaches for 3D-OCT, we do not rely on SNR or contrast in the underlying OCT data for performing motion correction, increasing versatility across different tissues types and OCT scanning approaches. By design, the technique does, however, assume that OCT intensity values are determined only by the optical backscattering of points sampled. Field curvature effects from the lens in the off-theshelf handheld OCT probe, used in this demonstration, therefore caused small artifacts (D) in the motion corrected images (see Figs. 6(f) and 8(d)). This effect could be readily removed by cropping pixels at the edges of the 7 mm × 7 mm uncorrected en face OCT image or by using an f-theta corrected telecentric lens. The use of external hardware, i.e., the in-built CCD camera, for motion tracking, also allows for the robust correction of large positional shifts. We demonstrate correction of up to 1.5 mm among B-scans (see Fig. 6(b)), limited primarily by the fiducial marker staying within the camera field of view. The ability to correct for hand motion is, however, limited by the 6 Hz sampling rate and a co-registration accuracy of within ∼16 μm, with higher frequency and small magnitude hand motion unable to be accounted for.
The use of a dual-function bi-layer is key to the robust correction demonstrated, enabling the decoupling of the accuracy of surface tracking and properties of the underlying tissue, since it removes the need to resolve features in photographs and OCT images. It also provides an easy visual check that the correction has succeeded. However, this is based on the assumption that the fiducial marker does not move relative to the tissue surface during the scan. The occurrence of such motion will cause incorrect alignment of the underlying tissue, as seen by the alignment artifact (A) in Fig. 7(b), since the photograph co-registration will favor aligning the fiducial markers. Thus, care must be taken to ensure that the bi-layer remains stationary with respect to the tissue during acquisition. We also note that the fiducial marker can present a potential disadvantage in reducing the effective field of view of OCT, optical palpation, and tissue surface photographs. Another issue for the compliant silicone layer used for optical palpation is the potential presence of negative stress, when the layer is compressed against uneven surface tissues. As the layer is incompressible, it has a tendency to expand into softer or cavernous areas e.g. the troughs of the fingerprint (see Fig. 7(e)), resulting in small negative stresses being calculated. More generally, surface topology can create stress contrast that may not be a direct consequence of the mechanical properties of the underlying tissue, as in the case of the fingerprint (see Fig. 7(e)). This could potentially obscure the desired contrast in optical palpation by complicating the direct relationship between stress and stiffness. However, higher fidelity stiffness contrast can be found by employing quantitative micro-elastography [35] and extending this motion-correction technique to it. Meanwhile, incorrect surface stress estimation in palpograms, obtained under the assumptions that the stress field within the compliant section of the bi-layer is uniform and uniaxial, and the interface friction is low could also be overcome in future work by extending previously demonstrated computational optical palpation to in vivo work [36].
Routine use of this technique in a clinical setting will likely require substantially faster volume acquisition. With the current approach of conveniently matching the camera frame and B-scan rates, we were limited to a maximum B-scan frequency of 6 Hz, but B-scan rates >10 kHz have been demonstrated [37]. The installation of a higher frame rate digital camera would enable this technique to be adapted to faster acquisition times. Alternatively, the technique could be adapted such that a photograph is captured every few B-scans, rather than every B-scan, and motion for B-scans between photographs could be estimated by interpolation. Since increasing the B-scan frequency increases the data acquired before detectable movement occurs, this would also eliminate much of the interpolation artifacts caused by insufficient sampling (as seen in Fig. 4(c)). Although much of the motion artifact demonstrated in this paper could be accounted for by a faster imaging system, it is important to consider that all scans were acquired by engineers who were consciously trying to hold the OCT probe stationary. It is anticipated that clinical scanning scenarios, such as in vivo scanning of a surgical cavity by a surgeon, will result in substantially increased bulk motion. Thus, even with faster scanning speeds, we anticipate that motion correction will play an important role in ensuring fidelity to tissue features.
Computation time also has important implications for the feasibility of our approach to handheld 3D-OCT and optical palpation in clinical applications, with the current processing time of a few hours requiring significant reduction. The majority of this time can be attributed to co-registering 807 photographs with the current off-the-shelf algorithm, chosen to enable proof-of-concept demonstration. Optimizing this photograph co-registration algorithm has the potential to greatly reduce this processing time, for example, by co-registering only a small window from each photograph or by choosing a feature tracking technique. This processing time could be also significantly reduced by re-implementing the code in C + + , which is much faster than the current Matlab implementation, and moving to GPU-accelerated processing [38].

Conclusion
Handheld imaging probes are needed to extend clinical applications of optical elastography to in vivo scenarios, yet are challenging to implement, largely because of motion artifact. In this paper, we present the first demonstration of optical palpation, a variant of optical elastography, with a handheld probe, without any form of supporting apparatus. Key to enabling this demonstration was the development of a dual-function bi-layer, providing both a fiducial marker for tracking and correcting for lateral hand motion, and a compliant section for estimating stress at the tissue surface. We also demonstrate, for the first time, that optical palpation can be used as a method for measuring pressure applied to the tissue during handheld operation, thus providing a more repeatable and robust imaging technique between different users. An off-the-shelf handheld OCT probe was used to successfully demonstrate this approach on a structured phantom, in vivo human skin and excised human breast tissue. All en face OCT images showed greatly reduced motion artifact, even when almost 1.5 mm of motion occurred during a scan, and optical palpation was shown to be valuable in contrasting features. Future work will involve integration with faster OCT scanning, incorporation of live applied pressure readings and extension for use in other optical elastography techniques. This work is an important step towards developing optical elastography towards routine clinical use.