2.1.

Tandem Interferometry

The conventional FF-OCT optical setup for microscopy is based on a Linnik interferometer.¹⁴ When considering its implementation as an endoscope, constraints of miniaturization, stability and resistance to environmental perturbations impose to divert from the classical configuration toward a design where the most exposed part—the probe—is as passive as possible. To do so, a tandem interferometry system is used, composed of:

The idea of cascading two interferometers was introduced in fiber-coupled low-coherence interferometry to prevent signal loss induced by polarization mismatch between the reference and signal beams.^15,16 In the present endoscopy configuration, the reference beam is the light back-reflected by the window at the tip of the probe and the signal beam is the light back-reflected by the sample in contact with this window within a slice of thickness $L_{\mathrm{C}}$, where $L_{\mathrm{C}}$ is the light source coherence length. The probe is thus insensitive to ambient perturbations since the two beams follow a common path through it. In the absence of a second interferometer, the maximum signal recorded would come from the slice of the sample in direct contact with the probe, corresponding to the zero path length difference. With the processing interferometer, a controlled path length difference is introduced, which needs to be compensated in the second interferometer so that the imaging plane is shifted further inside the sample.

2.2.

Full-Field Optical Coherence Tomography Endoscope

Figure 1(a) gives an overview of the endoscope prototype optical design. As depicted in Fig. 1(b), the system presents a gun shape, where the grip carries the light source and associated optics, the barrel is the optical probe, and the back that supports the camera is the imaging arm. With a diameter of 5 cm and a length of 15 cm, the grip allows easy handling of the 1-kg total weight device. The $20\mathrm{cm}(\text{length})\times 5\mathrm{mm}(\text{diameter})$ probe is long and thin enough to reach many of the endoscopic examination areas.

Fig. 1

(a) FF-OCT endoscope prototype setup and (b) pictures. ${\mathrm{L}}_1-{\mathrm{L}}_3$, lenses; BS, beam splitter; FPI, Fabry–Perot interferometer; PZT, piezoelectric actuator; PR, partially reflective.

A high-power LED (M660L3, Thorlabs) is used as a low temporal and spatial light source (${\lambda }_0=660\mathrm{nm}$, $\mathrm{\Delta }\lambda =25\mathrm{nm}$), with a pair of lenses to conjugate the LED chip with the entrance of the optical probe (L1: 6-mm diameter, 21-mm focal length; and L2: 9-mm diameter, 22-mm focal length). The processing interferometer is positioned between the two conjugation lenses, where the light is collimated. After passing the processing interferometer, the light is reflected by a beam splitter plate before being injected into the optical probe so that 50% of the light power is lost and another 50% of the signal power is lost on its way back to the camera. Contrary to conventional medical endoscopes with peripheral illumination and central signal collection, the beam splitter plate is crossed both by the injected light emitted by the LED source and by the light coming from the sample, which explains such a significant power loss and the need for a high-power light source. The optical power delivered to the sample is $1 \mathrm{mW}$. The probe is composed of a 20-cm long pitch 2 relay gradient index (GRIN) lens (GRINTECH, Germany) showing a diameter of 2 mm and a relatively small numerical aperture of 0.11 for image transfer, and of an objective lens (49271, Edmund Optics) with a focal length of 6 mm and a higher numerical aperture of 0.17 in order to ensure good lateral resolution. The probe ends with a planar partially reflective window (thickness 0.3 mm, diameter 2 mm) in order to create a surface where a plane contact is established with the tissue and where the reference wave originates. When in contact with biological tissue, the distal window reflects 8% of the incident light. This reference beam, as well as the signal beam coming from the imaging plane, which is the focal plane of the objective lens, travel back through the probe and are transmitted through the beam splitter plate, then conjugated with the camera ($1440\times 1440\text{pixels}$, $12\mu \mathrm{m}$ pixel side, 700 Hz maximum frame rate, 2 million electrons full well capacity, Adimec, the Netherlands) thanks to an achromatic doublet lens of focal length 75 mm and diameter 15 mm. The camera integrated in the FF-OCT endoscope is also a prototype, which has been developed jointly by CMOSIS (Belgium) and Adimec in the framework of the CAReIOCA European FP7 research project to meet the specific technical requirements of FF-OCT in terms of full well capacity and speed. Internal mechanical mounting elements are custom-made in order to minimize space loss. An external protective housing is made from light and thin plastics to reduce weight. Custom-made Labview software is used to control the instrument and the acquisition. In particular, the software synchronizes the camera triggering with the movement of the piezoelectric actuator located on the processing interferometer in order to perform phase shifting OCT detection.

2.3.

Processing Interferometer

In previous realizations of FF-OCT endoscopic setups using tandem interferometry, the processing interferometer is a Michelson-type one.^12,13 Because it is bulky, it has to be separated from the imaging probe in order to optimize the access possibilities of the device. This leads to the use of a multimode fiber to prolong the optical path between the two interferometers, which increases alignment issues and enhances the critical aspect of good parallelism tuning of the first interferometer. In our system, a low finesse, tunable Fabry–Perot interferometer is used (labeled FPI in Fig. 1) and easily inserted after the source and its collimating lens. Composed of two $50:50$ plate beam splitters of diameter 12.5 mm, one of them being attached to a piezoelectric actuator (PA 8/14, Piezosystem Jena, Germany), the Fabry–Perot interferometer extends on 25 mm only in the direction of light propagation and presents a section of $35\mathrm{mm}\times 45\mathrm{mm}$ transversally. The thickness of the air space between the two plates is precisely defined and controlled through the measurement of the fringes period with a spectrometer before integration into the endoscope. The thickness $e$ determines the depth of the slice imaged with the FF-OCT device, which is equal to $e/n$, where $n$ is the optical index of the sample (average 1.38 for biological tissue).

By changing the piezoelectric actuator offset voltage, the thickness of the Fabry–Perot interferometer can be modified, thus making it possible to scan the imaging plane through the sample depth and retrieve 3-D information. However, the depth of imaging is limited to the depth of focus of the optical probe¹³ defined by the objective lens and the wavelength used, here set at $50\mu \mathrm{m}$. For this first prototype endoscope, we did not include an axial scanning option, and the position of the imaging plane is fixed at $20\mu \mathrm{m}$ behind the surface of the tissue thanks to the Fabry–Perot interferometer thickness being tuned to $28\mu \mathrm{m}$ (fringes period of 7.5 nm). The piezoelectric actuator thus only provides the low amplitude modulation ($3.5{\mathrm{V}}_{\mathrm{PP}}$) needed to introduce a $\pi$ phase shift during one image acquisition in two. The displayed OCT signal is the difference between two consecutive images.

2.4.

Performance

The main characteristics and performance of the endoscope are summarized in Table 1. The transverse resolution and the field of view are measured experimentally by imaging a high-resolution 1951 USAF target, and the axial resolution is also experimentally evaluated by translating a mirror through the depth of focus and measuring the full width at half maximum of the fringe envelope. Compared to the system presented in Ref. 13, the transverse resolution evaluated at $2.7\mu \mathrm{m}$ is better with this endoscope thanks to the improvement of the numerical aperture of the distal optics, but the axial resolution of $6\mu \mathrm{m}$ is poorer because of the spectrum of the light source. It was decided for the light source to prioritize optical power over spectral width so that the imaging speed is higher (to minimize the influence of movements), but the axial resolution is altered. Moreover, this sectioning ability of $6\mu \mathrm{m}$ stays close to the typical thickness of histology slides, typically $4\mu \mathrm{m}$.

Table 1

FF-OCT endoscope prototype technical data.

The detection sensitivity is affected by the tandem interferometry configuration, as explained in Ref. 16. As a consequence, a 6-dB (four-fold) decrease in the SNR is theoretically expected in comparison with the traditional single Linnik interferometer.¹⁶ In order to evaluate the SNR loss in practice, a rat brain slice fixed in formalin is imaged ex vivo in several areas consecutively with the endoscope and with the LLTech commercial FF-OCT microscope (Light-CT scanner). All the experiments reported in this paper followed European Union and institutional guidelines for the care and use of laboratory animals. For the purposes of this comparative study as well as to evaluate the performance enhancement brought by the new camera, the endoscope is tested either with the Adimec camera or with the so-called conventional camera ($1024\times 1024\text{pixels}$, $10.6\mu \mathrm{m}$ pixel side, 140 Hz maximum frame rate, 0.2 million electrons full well capacity, Photon Focus, Switzerland), which is the camera classically used in FF-OCT setups. The result of the comparison on two different areas of the rat brain slice is given in Fig. 2. On the left, the images recorded with a Light-CT scanner are shown. In the middle and on the right, the images taken with the prototype endoscope equipped with the Adimec camera and with the conventional camera, respectively, are shown. The images have been resized and cut in order to limit the comparison to a common zone at a common scale. The conditions of illumination (camera saturated at 80%) and averaging (30 accumulations) during the acquisition are also the same for each device. A tolerance of about $10\mu \mathrm{m}$ should be considered regarding the adequacy of the imaging planes with the Light-CT scanner and the endoscope given that the axial resolution and the compression of the sample are not the same in the two experiments.

Fig. 2

Comparative study between (middle) FF-OCT images taken with the endoscope equipped with the Adimec camera or (right) with the conventional camera and (left) with the commercial Light-CT scanner. The sample is a rat brain coronal slice, imaged at $20\mu \mathrm{m}$ in depth with 30 averaging. The two rows correspond to two different areas of the sample.

The experiment shows a three-fold (5 dB) decrease in the SNR between the microscope and the endoscope equipped with the same camera. The possible error of adequacy between the two imaging planes, the difference of slice thickness due to the difference of axial resolution, and the existence of larger speckle grains on the endoscopic images, which reduces the possibility of finding uniform areas where the SNR can be compared with the Light-CT scanner images, are the possible reasons the image quality loss is lower than theoretically expected. The sensitivity of $-88\mathrm{dB}$ usually expected from the Light-CT scanner under 30 accumulations is thus reduced to $-83\mathrm{dB}$ with the endoscope prototype. In spite of the SNR alteration, anatomical features such as axons and myelin fibers or bundles are easily recognized on the endoscopic images, which is an important pillar of the diagnostic.

When the endoscope is equipped with the Adimec camera instead of the conventional camera, the SNR measured on the images increases by a factor of 2.7. This enhancement is the consequence of a higher full well capacity, provided that the light power is enough to saturate the sensor pixel wells. As the theoretical gain factor in full well capacity between the two cameras is 10, the expected gain factor in SNR is 3.2, since the SNR is proportional to the square root of the full well capacity (shot noise limited acquisition). As the Adimec camera is a prototype, the gain in full well capacity may be reduced, which explains the difference between the measurement and the theory. Another way to increase the SNR of the OCT images is to average more. Following this strategy, seven times more accumulations are needed to reach the same image quality with the conventional camera. Consequently, the second advantage of the Adimec camera is that at a given SNR, the imaging speed is multiplied by 4.