1. Introduction

Optical coherence tomography (OCT) is a well-established non-invasive, biomedical optical diagnostic imaging modality that enables in vivo cross-sectional tomographic and 3D visualization of internal microstructure and functional information in biological systems [1]. This technology is nowadays used for biomedical imaging and has found successful applications in ophthalmology, cardiology and dermatology, including numerous technology advances. When considering the key aspect for a successful transfer of a new biomedical imaging technology into clinical practice, the technology must provide not only the required biomedical information the investigator is looking for in a clear and easily understandable manner, but it must also include some practical and functional characteristics that justify the effort spent in learning, training and/or implementing the new technology. Typical requirements in biomedical imaging include real-time, non-invasive, in-vivo imaging with micron-scale isotropic resolution and sufficient penetration, short measuring time (especially when fast in-vivo functional biological processes are under investigation), simple and convenient interfacing of the device with the sample under test, integration of dedicated delivery probes and achievement of meaningful imaging results. Recent OCT technology is capable of fulfilling most of the requirements needed by lab development and/or clinical practice [1]. Following these guidelines for new technology adoption, from its beginnings [2] to date, OCT technology has rapidly improved and evolved from time-domain OCT (TD-OCT), to spectral/Fourier-domain OCT (SD/FD-OCT) and to swept-source OCT (SS-OCT). Advances in OCT technology went hand-in-hand with broadband light sources development. The light source represents a key technological element for OCT systems and the specific laser features directly define OCT system design and performance, specifically in terms of imaging speed and resolution. Various tunable source technologies are currently used in SS-OCT. Swept source OCT takes advantage of simultaneous high scanning speed and high axial resolution, together with the advantage of simple, cost-effective system and detector architecture [1], when compared with other OCT technologies, such as SD/FD-OCT. State-of-the-art swept sources are based on mechanical movements, e.g. Fabry-Perot tunable filter [3], MEMS tunable filters [4] or spinning mirrors [5] to achieve the desired sweep range and speed. New optically-pumped, externally-amplified MEMS tunable Vertical Cavity Surface Emitting Laser technology enables 100 nm tuning range at 1065 nm and 1310 nm central wavelength with single mode operation, resulting in several centimeters coherence length [6,7]. The mechanical movement, which represents the key aspect around which the above mentioned laser technology designs are based on, could also represent the limiting factor of the technology itself, limiting, by consequence, laser performance and imaging quality. Swept-wavelength laser technologies rely on highly repeatable, periodical and very stable sweep actuation; in simple terms, the more these criteria are met, the better the laser performs. All mechanical laser tuning mechanisms (such as tunable filters or MEMS mirrors) require accumulation and depletion of momentum for its completion, which can be subject to hysteresis (e.g. due to the mechanical movement) or experience undesired drifts of operational regime, e.g. due to external sources, such as thermal drift due to friction. Unstable or drifting mechanical movements can result in overall degradation of source performance. Other aspects that can influence swept laser performance are cavity length and laser optics design. Relatively long cavity length and complex laser source optics might result in cavity and/or mechanical instability (e.g. unwanted vibrations of the system), introducing another degradation mechanism in laser performance. In this paper we present – to the best of our knowledge for the first time – results from a novel akinetic, all-semiconductor swept laser source from Insight Photonic Solutions, Inc. (Lafayette, Colorado, USA). The innovative technology implemented in the laser involves no mechanically-moving parts (akinetic) to generate the sweep; the laser is based on integrated semiconductor opto-electronic design without the need of external cavity coupling. The all-semiconductor laser cavity is ~2 mm in length and is monolithically-constructed within the semiconductor. The all-semiconductor design enables a full electronic control of laser operation. The akinetic, all-semiconductor technological approach and design allows the akinetic laser to overcome most of the limitations encountered with mechanical sweep-based design implementations.

2. Methods

In this work, novel, unique, cost-effective, compact, all-electronic wavelength tuning light sources (Insight Photonic Solutions, Inc.; Lafayette, Colorado, USA) using a monolithic semiconductor device [8,9] that requires no moving parts (akinetic) or long fiber lengths are used for 2D and 3D SS-OCT in vivo and in vitro imaging at different scanning speeds, different center wavelengths and wavelength ranges. At 1550 nm, measurements are performed using two distinct laser sources with different optical bandwidths and hence axial resolutions. Measurements are also presented for a variety of conditions at a center wavelength of 1310 nm using a third distinct laser source. Laser performance related to OCT imaging is discussed, including roll-off, phase stability and relative intensity noise (RIN) analysis. A brief discussion concerning the aspects of the Insight laser source integration into OCT systems, including considerations on system design related to data acquisition and processing, is also presented.

2.1 Akinetic laser technology

The Insight laser technology uses semiconductor structures integrated within a single laser chip for wavelength tuning, therefore eliminating the need for mechanical tuning elements, as traditionally used in state-of-the-art OCT swept sources. The parameters that define laser performance are software-controlled and the most relevant of them, e.g. sweep rate, sweep tuning range, output power and spectral output profile, are accessible to and modifiable by the user. This gives the laser a unique flexibility to adapt to specific user requirements and imaging performance necessities. The sweep tuning mechanisms of the all-semiconductor laser are based on controlled changes of the refractive index of the semiconductor materials composing the laser cavity. The semiconductors’ refractive index value is an expression of the electronic current flowing through it [10]; a precise control and synchronization of the currents flowing through the cavity semiconductor elements yields precise control of laser tuning.

Figure 1 shows a cross-sectional schematic representation of the laser cavity. The representation illustrates five semiconductor tuning elements with associated current source controller: two Distributed Bragg Reflector (DBR) mirrors (elements a and d); a cavity gain medium unit (element b), which provides the laser cavity amplification; a cavity length adjustment unit (element c); and a semiconductor optical amplifier (SOA) unit (element e). The mirrors provide wavelength dependent feedback to the selected lasing wavelength; by applying a desired current, a change in the refractive index of the two mirror elements is induced – i.e. changing the preferred wavelength of the lasing radiation inside the cavity. Based on the Vernier-tuning effect, control over the currents enables selecting any lasing wavelength across the tuning range. The cavity length adjustment unit (element c) allows additional control of the effective optical path length between the two mirrors, enabling real-time fine adjustment of cavity length and single mode operation. The SOA provides final boosting and spectral shaping of the output power of the laser. The control and synchronization of the electronic currents is software actuated by means of a field programmable gate array (FPGA). The implemented control algorithms constantly monitor and adjust current intensities and phases to ensure correct, periodic and repeatable laser sweep. One direct benefit of the implementation of the programmable controlled Vernier-tuned distributed Bragg reflector (VT-DBR) is that the laser is mode hop free; long before the akinetic laser could even begin a mode hop, the control algorithms intervene and modify the sweep parameters to ensure optimal side mode suppression ratio (SMSR) and guaranteeing no mode hops. SMSR is a measure of the ratio of the peak amplitude of the primary single longitudinal mode of the laser relative to the amplitude peak of the nearest side mode of the laser, or, in the absence of a side mode, of the highest feature of the background ASE spectrum, expressed in dB. Resolution, sensitivity and general image quality obtainable by an OCT system are influenced by many factors, of which the source SMSR is one part. OCT image characteristics are typically evaluated in the domain of the point-spread function (PSF), by measuring the full-width half-maximum (FWHM) of the PSF – a common measure of image resolution; the ratio between the PSF peak amplitude relative to the nearest PSF side-lobe peak amplitude – an estimation of image contrast or dynamic range achievable in a system; and PSF peak-to-noise-floor ratio – an estimation of system sensitivity (noise floor measured with the sample arm of the interferometer blocked). If a secondary mode is present in the laser cavity, it can introduce unwanted additional features in the PSF, due to the interference of the secondary mode with itself and with the primary mode. These features may appear as side-lobes in the PSF, reduced in amplitude by the SMSR (primary-secondary mode interference) or by squared SMSR (secondary-secondary mode interference) and aliased into the PSF frequency domain, below the Nyquist sampling limit of the system. The aliasing effect introduced by the secondary mode may appear throughout the PSF signal, affecting the overall noise floor of the PSF. If the secondary modes introduce side lobes near the PSF main peak, this can clearly degrade edge contrast and even reduce system resolution by broadening the PSF’s FWHM. When the secondary modes of the laser interfere with the primary and secondary modes, the interferences will alias back into the desired OCT signal bandwidth and increase the noise floor. This is true even when the sample arm is blocked, since the primary and secondary modes may still interfere along the reference path as an unintended common path interferometer. Hence, secondary modes of the laser may reduce the dynamic range of an image as well as reduce the sensitivity of the image. The presence of secondary modes may also induce residual mode competition which introduces amplitude fluctuations of the laser, which is associated to RIN, yielding reduction in sensitivity. If and where this becomes an issue varies for each laser source and imaging system. Systems with better balanced photodetection and common mode cancellation should be less sensitive to increases in source RIN. Nonetheless, if a secondary laser mode increases the source RIN, then that will reduce system sensitivity. The akinetic laser has multiple optical and electronic feedback mechanisms to optimize sweep performance that directly influence the decisional processes of the optimization algorithms. Adjustments actuated by the control algorithms on the internal parameters which control the sweep aim to increase both the primary laser mode’s amplitude and reduce the secondary laser mode’s amplitude. Depending on specific sweep states, the laser may require adjustments that increase the primary mode more, or reduce the secondary mode more, or the algorithms may do both, trying always to find the optimum balance between the two. The programmable controlled VT-DBR also enables all-electronic single longitudinal mode operation and tight control over the optical characteristics [8,9]. Optical state changes in the cavity, i.e. transitions between any wavelength to an adjacent one as a result of a step-change in the control currents, are timed by a 400 MHz master clock, allowing any new optical state to be generated from the previous every 2.5 ns. The optical states transition time imposed by the master clock is a constant of the laser source and directly influences laser operability. The laser can generate up to 400,000,000 wavelength points per second, which yield up to e.g. roughly 2,000 points at a 200 kHz sweep rate. Not all the 2,000 points are associated to meaningful wavelength values (or cavity optical states) composing the final sweep as required for OCT imaging, though. This is a consequence of the implemented technology, and has a direct influence on general SS-OCT system design and on how acquired (digitized) data is processed. More specifically, the laser sweep rate is user-selected by choosing the number of points (i.e. number of wavelength values) which build up the final spectrum during one sweep, e.g. 1550 points need to be chosen to generate 200 kHz sweep rate over 40 nm sweep range for the 1550 nm model – ref. Table 1 in section 2.3. In building up the wavelength sweep, the laser does not generate a subsequent and contiguous ensemble of valid cavity optical states in one single pass, to fully cover the whole selected sweep range. Instead, inside the laser, the sweep is built by stitching sweep sub-intervals (valid points) interspersed by transition intervals (invalid or discarded points), imposed by the laser technology and necessary to update the driving currents status. During the transition intervals, which can last several master clock ticks, the optical state of the laser cavity is changing, reflecting the associated values assumed by the control currents. These optical states are finite but undefined – in the sense of useful contribution to build up the spectrum for OCT imaging – and should not actively contribute to the formation of the final output spectrum. When the laser is actively sweeping, the cavity optical state is constantly changing and updating every 2.5 ns, and each new state is available – always and simultaneously – to the laser output. The optical state assumed by the cavity at each instant, always reflects the values expressed by the driving currents at the same time, making no difference if the driving currents are intended to generate a valid optical state (valid point or wavelength composing the sweep) or performing a transition (invalid points) between valid sub-intervals. Since the output is not disabled during the transition interval time slots, the final expression of the output spectrum obtainable from the Insight laser source after one sweep is given by a time-sequenced ensemble of valid wavelength values and invalid wavelength values, the latter generated in the laser cavity during the transition interval time slots and sent to the output. The internal processes of the laser are responsible to automatically increase the user selected sweep points (e.g. user-selected 1550 points to generate 200 kHz sweep rate) to a larger number to include the necessary and unavoidable steps (e.g. changes in the driving currents) taken during the transition intervals. The transition time intervals may require several master clock ticks to move from one valid subset of sweep wavelength to the next, producing undesired optical states at the output of the laser at each transition tick. The number of ticks (i.e. time duration) required by each transition are mainly expression of the time required by the laser’s electronic circuitry to set the correct changes in the driving currents values, on the parameters values applied to the control algorithms and to the selected sweep rate. Typically, a total of about 450 transition or invalid points (the sum of all transition intervals during one single sweep) are added by the laser to the user-selected 1550 sweep points to produce a sweep length that is about 2000 total points long, yielding a sweep repetition rate of 200 kHz. The number of unwanted points may vary by changing the values of internal (laser) parameters adjusted in the calibration algorithms; by the quality of the calibration process; or by the selected sweep rate (over a fixed sweep range, the number of valid sweep points defines the wavelength step between sweep points). The determination of when transitions occur in a sweep, and how long the transitions last, is made by the laser source and is not directly controllable by the user. The ratio between valid points (user selected) and total points (valid + invalid) per sweep also defines the laser’s sweep efficiency, which can be different for different selected sweep rates (see section 2.2). For a correct reconstruction of the sweep spectrum/SS-OCT signal, the invalid points in the output sequence need to be removed; this is typically done in post-processing, just after the OCT signal has been digitized and before the application of any other data processing. Due to stable and repeatable laser operation, the positions of valid and invalid points in the acquired raw data sets are constant in every sweep and change only when the user requests a new laser sweep calibration. The information describing the position of the valid and invalid points within the sweep is available to the user via two methods. First, the positions of valid and invalid points are generated by the laser during the calibration process, stored in a numeric array and available to the user for later use. Alternatively, the valid data information can also be accessed in real time at each sweep from a 400 MHz low-voltage TTL signal (named Data Valid Trigger) available as laser output signal and synchronized with the sweep. The Data Valid Trigger is associated with and synchronized to the time-sequenced laser sweep. Each TTL value defines the validity or invalidity (e.g. 1 – valid; 0 – invalid) property of the correspondent (simultaneous) value composing the laser’s wavelength sweep. The described method for generating a sweep spectrum for OCT imaging impacts the way the signal from the OCT interferometer must be acquired and processed, and will be discussed in section 2.6. The minimum and maximum sweep wavelengths can also be selected by the user (option available in future versions of the laser firmware), up to the lowest and highest wavelengths supported by the laser chips.

 

Fig. 1 Schematic representation of Insight laser cavity.

Download Full Size | PDF

Table 1. Main specifications of Insight swept laser sources as used in the experiments. λ₀ – spectrum central wavelength; Δλ – sweeping range; P_out – laser output power; Δz – system axial resolution (depth), in air.

View Table | View all tables in this article

2.2 Duty cycle vs. sweep efficiency

The introduction of invalid points during active sweep requires some refinement in the definition of the duty cycle when applied to the akinetic all-semiconductor laser technology. Duty cycle for swept sources is generally defined as the ratio of the sweep time to the sweep repetition period, where sweep time is understood to be a continuous time interval over which the generation of the output wavelength sweep is taking place; and sweep repetition period is the sum of the sweep time and the time spent by the swept laser source to (re)create the internal initial conditions (laser state) to generate the next wavelength sweep and eventually to synchronize with the acquisition hardware. The time for the laser to return to the initial conditions to repeat a sweep is often referred to as “fly-back time”, alluding to the mechanical nature of the tuning. Sweep efficiency is defined as the ratio of the sweep time portion that produces useful data (i.e. generates useful contribution to the measurements) to the total sweep time. For most of the mechanically-tuned swept sources, any and all instants of the sweep time contribute to the formation of the (useful) spectrum, and the localization of the sweep time and fly-back time within the sweep repetition period is well-defined; the fly-back time is finite and limited by the laser’s mechanical tuning elements. Sweep efficiency describes the fraction of the sweeping time that produces useful data. For a mechanically-tuned laser, the sweep efficiency may be 100%, since the sweep may be sampled 100% of the time during an OCT measurement. But for the same laser, the duty cycle might be 40-70%, due to the non-negligible finite duration of the fly-back time. In the akinetic all-semiconductor Insight lasers, the situation is somewhat reversed. Sweep efficiency of the akinetic lasers varies from ~65% up to ~96%, depending on the sweep rate. Conversely, the duty cycle may be over 99%, since the end of the sweep of the laser may be quickly transitioned to the beginning of the sweep. The number of points, or the amount of clock ticks, spent during the fly-back can be software controlled and arbitrary chosen. This feature is also user-selectable by specifying the duty cycle value, such that the sweep control algorithms may add transition points to the end of a sweep to lower the duty cycle; this may be a useful feature for a number of applications or specific hardware that requires some specific time to be spent between sweeps to, e.g. re-arm the ADCs. Figure 2 shows some examples of swept laser sources duty cycles/sweep efficiency implementations.

 

Fig. 2 Graphic representation of the duty cycle and sweep efficiency for (a) mechanically-tuned swept laser; (b) VT-DBR akinetic all semiconductor laser. The black curves represent the laser generated optical frequency (or associated wavelength) at each instant (reading). The sweep repetition time interval and related sub-intervals definitions are color-coded and overlapped to the frequency curves. Typically, for (a), the sweep time and the “valid points” time intervals coincide (100% sweep efficiency). (c) illustrates the time sub-intervals distribution along one sweep for the akinetic laser source, highlighting the distribution of valid (green background) and invalid (orange background) time slots.

Download Full Size | PDF

2.3 Experimental setup

Figure 3 shows a schematic representation of the imaging engine as used in the experiments. The system implemented a standard Michelson interferometer scheme with dual-balanced photodetection. The laser output was coupled into port 1 of a circulator (cr₁) and delivered into the 50/50 single mode optical coupler (50/50) via port 2 of cr₁. There the probing radiation was split according to the coupler’s splitting ratio – i.e. equally split, in our system – and diverted towards the reference and sample arm respectively via collimating optics (C).

 

Fig. 3 Schematic representation of the imaging setup as used in the experiments. laser – Insight laser source; cr₁, cr₂ – circulator; 50/50 – fiber-based single mode optical coupler, 50/50 splitting ratio; p – polarization controller; c – collimator; R – retroreflector; M – mirror; xy scan – 2-axis galvo scan unit; L – imaging lens; PD – dual-balanced photodetector; DAQ – digitizer; PC – personal computer. See text for details.

Download Full Size | PDF

The back-reflected (reference) and backscattered (sample) portion of the light coupled back into the optic fiber, crossed back through the 50/50 coupler, where it split and interfered. The split outputs from the coupler travelled through ports 3 of the respective circulator (cr₁ and cr₂) and were directed towards the dual-balanced photodetector (PD, Exalos AG, Schlieren, Switzerland; mod. EBR Balanced Receiver). The converted optic-to-electric interferometric signal was first low-pass filtered with a 190 MHz LP filter (Minicircuits, mod. SLP-200 + , not shown in Fig. 3) and then acquired by the 12-bit A/D digitizer (DAQ, Alazartech Technologies, Inc., Canada, mod. ATS9350) at 400 MS/s and sent to the personal computer (PC) to be converted into OCT image. The DAQ sample clock signal was derived from the Insight laser Sample Clock output, to synchronize laser sweep (generation of new cavity optical state or wavelength) with A-scans sample acquisition. The second circulator (cr₂) was inserted in the system to create (as much as possible) symmetrical optical paths to be crossed by the interfered probing radiation while back-propagating from the 50/50 coupler towards the photodetectors.

2.4 Sources characterization and performance

Three different prototypes of the Insight akinetic all-semiconductor swept laser source were evaluated and used for high-speed SS-OCT imaging. Results from two swept laser sources at 1550 nm central wavelength and different sweep range (Δλ), and from one at 1310 nm central wavelength were obtained. Table 1 summarizes the main specifications of these sources as used in the experiments.

Based on the imaging engine schematic as illustrated in Fig. 3, two similar imaging systems were built for the three sources; one optimized for the 1550 nm central wavelength light sources and the other optimized for the 1310 nm central wavelength light source.

The two systems shared the 2-axis galvo unit (xy scan, Thorlabs, mod. GVS002) and the optic elements of the reference arm, and differentiate from each other in the fiber-based and remaining free-space components, which were chosen to optimize the system’s performance according to the used laser’s central wavelength (1310 nm or 1550 nm). The observed output power spectrum profiles for the three laser sources were extremely flat (< 0.015 dB) along the whole sweeping range, e.g. we observed contained variations – within 0.015 dB – of the output power spectra along the whole sweep range of the 1550 nm, 40 nm bandwidth, 5.5 mW Insight laser source, as shown in Fig. 4. The sharp boundaries of the output power levels that can be observed in Fig. 4 are a direct consequence of the programmable nature of the laser: the source does not tune outside the software programmed wavelength limits, therefore generating zero output power outside the selected sweeping range. The sweep repetition rate was adjustable from 7 to 200 kHz with >75% duty cycle, since this all-electronic tunable swept sources moved from any optical state to any other optical state in 2.5 ns. The OCT system’s probing beam spot size (diameter) was ~20 µm, measured in the sample arm’s focal plane. OCT system sensitivity >95 dB was accomplished with the 1550 nm sources and >90 dB for the 1310 nm source. Sensitivity values were acquired at 100 kHz sweep rate, with ~2 mW and ~0.5 mW incident power on the sample for the 1550 nm and 1310 nm laser source respectively. Further improvement is expected with improved detection and power levels from the source approaching 20 mW. The side mode suppression of the akinetic swept source was at least an order of magnitude better than that of other commercial lasers [11]. Most of the commercially available swept lasers offer (double pass) coherence lengths (@ −6 dB signal drop) between 3 and 15 mm, with one expensive polygon mirror unit that provides coherence lengths >50 mm, and an optically-pumped, externally amplified MEMS tunable VCSEL laser with coherence length of several centimeters [4,12,13]. The cavity of the akinetic swept source is ~2 mm long with a single internal longitudinal mode, therefore inherently offering long coherence length. In our SS-OCT setup employing akinetic swept sources, OCT imaging at 1550 nm was achieved with more than 170 mm double pass coherence gate displacement, with −1 dB signal drop within the first 40 mm and no significant changes between 20 and 200 kHz sweep speeds (see Fig. 5 and Fig. 12). To estimate laser performance, point spread function (PSF) roll-off curves, laser source phase stability and relative intensity noise (RIN) were evaluated. PSF peak value remained approximately constant for most of the displacement range, observing instead a rise in the noise floor at higher distances (>20 mm) from the zero delay.

 

Fig. 4 Normalized output power spectrum (flat shape) of Insight 1550 nm 40 nm sweep range, 5.5 mW output power swept laser source. The inset shows a magnification of a portion of the power spectrum (valid and invalid points).

Download Full Size | PDF

 

Fig. 5 PSF roll-off curves of Insight 1550 nm 40 nm sweep range laser source recorded at (a) 20 kHz, (b) 100 kHz and (c) 200 kHz sweep rate. Plots (a), (b) and (c) show the PSF decay in the 0-180 mm, 0-40 mm and 0-20 mm depth range respectively, which represent the whole depth range allowed by the mechanical limitations of the reference arm for (a) and the associated selected sweep rates for all. The PSF peak values drop-off and the appearance of associated side lobes (some pointed by the black arrows as indications) were likely related to Nyquist limitations.

Download Full Size | PDF

Figure 5 shows the roll-off plots for the 1550 nm 40 nm sweep range Insight laser source recorded at 20 kHz, 100 kHz and 200 kHz sweep rate respectively. To evaluate the system’s and laser’s sweep phase stability at different sweep rates, auto- and cross- correlation measurements were performed. For both measurement types and for each selected sweep rate (20 kHz, 100 kHz and 200 kHz) a set of ~32000 consecutive spectra (sweeps) were collected from the same point on the sample (no xy scan active) using a 1 mm thick cover glass as sample in the sample arm, with the sample arm focal plane positioned inside the cover glass just beyond the front surface. Figure 6 reports a schematic representation of the algorithms applied to the acquired spectrums (raw data) for both cross- and auto-correlation phase stability analysis. Auto-correlation analysis intended to estimate the laser’s sweep phase repeatability; in this case, data were obtained with the reference arm signal blocked while recording the self-interfering signal (front surface of the cover glass referred to its back surface) generated from the sample arm. Assuming that the self-interfered signal did not experience significant disturbances in the phase when travelling outside the laser source, the computed phase variations with the auto-correlation analysis (auto-correlation branch in Fig. 6) are attributed mainly to overall phase variations arising inside the laser source. Cross-correlation analysis intended to estimate the laser phase linearity and phase stability of the system (laser + interferometer), which represents useful information in those cases where OCT imaging is obtained from the phase of the interferometric signal, such as Doppler OCT imaging. In this case measurements were conducted with the reference arm unblocked (conventional OCT imaging) while recording the interferometric signal. Standard deviation of phase differences of the recorded data evaluated at the PSF peak position coincident with the front surface of the cover glass were computed at the three sweep rates following the procedure illustrated in Fig. 6, “cross-correlation” branch.

 

Fig. 6 Schematic representation of the algorithms applied to evaluate the laser (auto-correlation) and system (cross-correlation) phase stability. S_i represents the i-th spectrum of the ensemble (raw data); ℱ{} the FFT operator; |•| the module operator; ∠ the phase operator; ℱ⁻¹{} the inverse FFT; $\widehat{F}$ indicates the filtered PSF; φ the unwrapped phase; φ_D the phase difference of consecutive spectrums; $\overline{\phi }$ the mean phase of the ensemble; $\overline{\phi }$_fit the linear fit of the mean phase; σ(•) the standard deviation computation. See text for details.

Download Full Size | PDF

These calculations gave an indication of laser (auto) and system (cross) phase jitter (φ) and displacement sensitivity (Δφ). In particular, σ_Δφ represents σ(φ_D,i) shown in the cross-correlation branch of the algorithm schematic in Fig. 6, evaluated at the PSF peak position relative to the cover glass front surface. Following computational steps along the auto-correlation arm of the algorithm of Fig. 6, phases of each recorded power spectrum evaluated at the cover glass front surface, φ_i, were extracted from the raw data. Each φ_i was computed by unwrapping the phase of the inverse FFT signal, being previously filtered (peak extraction) to isolate the PSF values at the sample’s front surface. The digital filtering operation was obtained by multiplying the computed FFT of the recorded power spectrums (F_Si) with a rectangular digital filter centered on the PSF peak and applying an additional Hanning window to the extracted portion of the FFT in order to reduce computational artifacts, e.g. induced oscillations due to Gibbs phenomena introduced by the application of the digital rectangular filter. The rectangular filter was defined as an array of zeros, except at the index position coinciding with the detected sample’s front surface PSF peak index position (derived from |F_Si|) and few neighbor samples, where it assumed the value of 1. In our computation, the selected digital rectangular filter length was 20 samples, symmetrically centered on the PSF peak index position (i.e. 10 points before and 10 points after the PSF peak index).

From the collection of unwrapped phases, the mean phase,$\overline{\phi }$, was extracted and used as a reference curve to evaluate the laser’s sweep phase repeatability, quantified by computing the standard deviation of φ_i,−$\overline{\phi }$, i = 1, … N, where N indicates the total number of occurrences (i.e. recorded spectrums) used in the calculations. A linear fit of $\overline{\phi }$, named $\overline{\phi }$_fit, was also computed and used as a reference curve for the laser’s sweep phase linearity evaluation; also in this case, standard deviation computation of φ_i−${\overline{\phi }}_{fit}$, i = 1, … N, was applied to quantify the sweep linearity. Figure 7 reports the measured average phases per sweep (Fig. 7(a)) and quality figures of phase linearity (Fig. 7(c)) and repeatability (Fig. 7(d)) of the Insight 1550 nm, 40 nm sweep range laser source, evaluated at the three sweep rates of 20 kHz, 100 kHz and 200 kHz. In all three cases, the computed laser’s mean sweep phases exhibited, with a very good approximation, linear behavior. Sweep linearity is illustrated in Fig. 7(a), where the three plotted curves of $\overline{\phi }$ for the three selected sweep rates are quasi overlapping to each other. Figure 7(b) shows plots of the difference between the mean phase and its associated linear fit curve for the three sweep rates, where it is possible to appreciate a contained variation of the phase of ± 0.2 rad over a span of ~335 rad per sweep (<0.06%) along almost the whole sweep range. Since the phase of a single sweep differs from its linear approximation by less than 0.06%, linearity and repeatability estimations offered very similar results (Figs. 7(c) and 7(d)). What can be observed from Figs. 7(c) and 7(d) is that an increase of the laser sweep rate yields an increase in phase variations (larger standard deviation) during the sweeps, degrading or altering linearity and repeatability. Our calculations revealed a standard deviation of the order of milliradians over a span of ~335 rad per sweep. We observed larger overall variations <2 mrad for the worst case at higher sweep rate, with the only exception of the observed peak of ~8 mrad at ~1558 nm wavelength for the 200 kHz sweep rate case. The presence of the ~8 mrad standard deviation peak in Fig. 7(c) is a consequence of the laser calibration process, whose performance is governed by the laser internal algorithms. It is expected these anomalies will be reduced or eliminated with improved versions of laser firmware. These plots are compared with the minimum detectable phase (difference) model valid under shot noise limit conditions [14]:

 

Fig. 7 Estimation of phase stability for the Insight 1550 nm, 40 nm sweep range laser source at 20 kHz (blue curves), 100 kHz (red curves) and 200 kHz (black curves) sweep rates. a)$\overline{\phi }$ = φ_mean mean phase, unwrapped; b) difference between mean unwrapped phase and associated linear fit curve ${\overline{\phi }}_{fit}$ = φ_fit; c) evaluation of sweep linearity: standard deviation of the differences between single sweep phases, φ_i, and${\overline{\phi }}_{fit}$; and d) evaluation of sweep repeatability: standard deviation of the differences between single sweep phases φ_i, and $\overline{\phi }$.

Download Full Size | PDF

Table 2. Estimation of phase (σ_φ), phase differences of consecutive spectrums (σ_Δφ) and phase under shot noise conditions (σ_sn, model) of power spectrum FFTs of auto-correlation (left side) and cross-correlation (right side) data sets using the Insight 1550 nm, 40 nm sweep range laser source. FFT phases are evaluated at cover glass front surface (PSF front peak position). The cover glass is 1 mm thick.

View Table | View all tables in this article

Auto-correlation shot-noise limited phase sensitivity measurements were obtained without the need for any external wavelength references, such as a temperature-stabilized Bragg grating or a gas cell [15]. Estimations of standard deviation of phase (σ_φ, σ_φD), phase differences (σ_Δφ, σ_ΔφD) between consecutive spectrums and phase under shot noise conditions (σ_sn) of the computed power spectrum FFT were referred to the cover glass front surface (PSF front peak position) for the cross- correlation data sets and to the cover glass front surface referred to cover glass back surface for the auto-correlation data sets. Direct application of the FFT algorithm on the recorded interference pattern (raw data) yields Fourier-transform limited peak widths. The slight deviations observable at the boundaries of the sweep (e.g. ref. Figure 7(b)) do not cause any observable broadening of the computed PSF, ascribing instead their presence mainly to the applied digital processing. Figure 8 shows the computed FFT for the auto-correlation (Fig. 8(a)) and cross-correlation (Fig. 8(b)) data set, acquired at 200 kHz.

 

Fig. 8 Direct FFT computation of acquired spectra (raw data) for (a) auto-correlation and (b) cross-correlation data sets. The peaks in (b) relate, from left to right, to the front and back surface of the cover glass respectively. Data acquired with Insight 1550 nm, 40 nm sweep range at 200 kHz sweep rate.

Download Full Size | PDF

Similar quality figures were obtained from the 20 kHz and 100 kHz data sets. The auto-correlation PSF peak shown in Fig. 8(a) is broadened due to dispersion experienced by the probing beam while crossing the 1 mm cover glass. Figure 8(b) shows two cross-correlation interferometric peaks, associated (from left to right) to the front and the back surface of the cover glass. The front surface peak (left peak in Fig. 8(b)) is composed by one single sample, indicating that the PSF is Fourier-transform limited. This is an expression of simultaneous good laser phase linearity and that the system is, with good approximation, dispersion-free. The cover glass back surface peak (right peak in Fig. 8(b)) is broadened due to dispersion introduced by the sample. A direct comparison of the shape (envelope) of this peak with the auto-correlation peak shape (envelope) suggests that the main contribution to peak broadening is originated by dispersion introduced by the sample and not by the system or by non-linear phase changes introduced by the laser.

2.5 Relative intensity noise

Relative intensity noise (RIN) is the measure of undesired fluctuations in the laser output power. The laser output power can be expressed as P = P₀ + ΔP(t), where P₀ represents the average optical output power, and ΔP(t) represents the time-dependent fluctuation with zero mean, i.e. the time-averaging value of the fluctuation is zero. The power of the fluctuation is characterized via the mean of its squared deviation, which can be expressed in the power spectral density function S_ΔP(f) [16]. We measured the RIN of the 1550 nm, 40 nm sweep range, 5.5 mW output power akinetic laser at different laser sweep rates. In our analysis, the power spectral density (PSD) function was calculated by computing the Fourier transformation of normalized recorded time series of power fluctuations. The measurements were performed for the 20 kHz, 100 kHz and 200 kHz laser sweep rates and with no active sweep. Power fluctuations traces of ~1.5 million points and ~3.5 million points for the 200 kHz and 100 kHz cases respectively, and ~19 million points for the 20 kHz and no-sweep cases respectively were Fourier transformed, and the square of the amplitudes displayed as power spectrum. The analysis also included the laser non-sweeping regime to estimate the amplified spontaneous emission (ASE), regarded as reference background noise, generated by the system and captured by the digitizer.

The spectrums were normalized to units of dBc/Hz by adding a constant on the logarithmic scale such that the DC component assumed the value of 10·log(1Hz/RBW), with RBW being the equivalent RF-resolution bandwidth defined as digitizer sampling frequency/trace length. RBW values of ~250 Hz, ~110 Hz and ~20 Hz were obtained for the 20 kHz, 100 kHz and 200 kHz sweep rates respectively. During the measurements, the laser output power was set to 5.5 mW, and a fraction of it, ~110 µW, was sent to the photodetector. Figure 9 shows the normalized power spectra of the laser during ASE (9(a)) and sweeping regimes (9(d), 9(g), 9(h)) at the different sweep rates. The ASE shows an approximately linear increase of the noise floor over the detection bandwidth; similar result was also obtained when no laser output was connected to the photodetector, therefore ascribing the cause of the rising of the noise floor due mainly to the photodetector. Figure 9(b) and Fig. 9(c) show a magnification of the PSD in ASE regime. With the laser connected to the PD and in sweep regime, the computed PSD show an increase in the noise floor level while approximately preserving a linear variation along the detection bandwidth, with the exception of the very low frequency region where a slight non-linear increase is observed. Figure 9(d) shows the computed PSD for the 20 kHz sweep rate; Fig. 9(e) and Fig. 9(f) show a magnified portion of the PSD at 20 kHz. The overlapped red traces in Fig. 9(d)-9(f) represent previously published data obtained with another OCT swept laser (ref [17].), regarded as the best term of comparison. It is interesting to notice that fluctuations associated with sweep repetition rate of the akinetic laser do not appear in the PSD data, unlike what can be observed in mechanically tuned swept sources. The three peaks observed in the PSD curves for the three sweep rates are originated by the calibration process (particularly the peak observed at ~80 MHz) and other laser internal processes. Similarly as commented for the phase stability analysis, these processes are software controlled and the reduction or elimination of these peaks might be expected with improved releases of laser internal firmware. The RIN analysis was based on the definition of sliding RIN and ortho RIN as from [17], which can explicitly express the relative intensity noise of the swept laser as a function of the wavelength. A detailed description of the definitions and methods that defines the sliding and ortho RIN is reported in [17]. In the case of the Insight swept laser, the wavelength-sampled point sweep relation is obviously straightforward (see section 2.1 and section 2.6), making the definition of Biedermann et al. an ideal evaluation method for the akinetic all-semiconductor laser types.

 

Fig. 9 Normalized RF PSD of Insight 1550 nm, 40 nm sweep range. (a-c) ASE with no active sweep; (d-f) PSD at 20 kHz sweep rate; (g) PSD at 100 kHz sweep rate; (h) PSD at 200 kHz sweep rate. Red traces in d-f indicate previously published data for a FDML swept laser source (ref [17].). The computed detector’s shot noise limit (~110 µW incident power) was −145.8 dBc/Hz.

Download Full Size | PDF

Figure 10 summarizes the results for sliding RIN and ortho RIN analysis, expressed in percentage. For the sliding RIN estimation, which is an expression of the “intra-sweep” noise, windows size of 179, 33 and 15 points were selected for the 20 kHz, 100 kHz and 200 kHz sweep rate respectively, all the cases yielding 1% wavelength variations within the window. The frequency range included in the sliding RIN moved from the system maximum bandwidth, 190 MHz in our case, down to 2f_SR/n, with n being the number of neighboring samples (window size) and f_SR the digitizer sampling rate (400 MS/s in our case). The frequency ranges for the 20 kHz, 100 kHz and 200 kHz sweep rates were ~185 MHz (from ~5 MHz to 190 MHz), ~166 MHz (from ~24 MHz to 190 MHz) and ~137 MHz (from ~53 MHz to 190 MHz) respectively. With reference to Fig. 10(a), the observed increase in standard deviation values at ~1557 nm and ~1562 nm wavelength for the 200 kHz curve is related to the calibration process controlled by the laser firmware and it is expected that it can be reduced or eliminated with improved versions of the calibration algorithms. The ortho RIN analysis, regarded as “inter-sweep noise” since it estimates power spectral fluctuations of several subsequent sweeps at each fixed wavelength, covers a frequency span in the RF bandwidth region from half the laser repetition rate, down to the inverse of the time interval required to acquire the sequence of samples at each fixed wavelength (i.e. DAQ sampling rate (time) × nr. of samples per sweep (valid + invalid) × nr. of acquired spectra). The temporal spacing for 20 kHz, 100 kHz and 200 kHz sweep rates for sequences of 1024 values (i.e. 1024 recorded spectrums) was respectively ~50 µs, ~10 µs and ~5 µs, yielding bandwidth ranges of ~20 Hz up to 10 kHz; ~100 Hz up to 50 kHz; and ~198 Hz up to 100 kHz for the three sweep rates respectively. The ortho RIN was calculated acquiring 1024 records (sweeps) of 19000, 3500 and 1550 samples (valid points) at 20 kHz, 100 kHz and 200 kHz sweep rate respectively. Per each data set, the standard deviation at fixed positions, i.e. fixed wavelength, within the record was calculated, yielding the wavelength dependence of the RIN δP/P(λ). The calculated RIN values are then the integrated noise spectral densities along the associated bandwidth range according to the selected sweep rate. The computed detector’s shot noise limit for our system was −143.8 dBc/Hz, with ~110 µW incident power on the detector – a fraction of the 5.5 mW output power from the laser. The ortho and sliding RIN curves in Fig. 10, for the three sweep rates and over all wavelengths, show a typical value of 0.15% that translates to a mean power spectral density of −139 dBc/Hz over the 190 MHz bandwidth of our system. The calculated shot noise limit for the akinetic laser at 5.5 mW is −163.3 dBc/Hz. Although the sweep ranges of the two compared sources are different (100 nm for [17] vs. 40 nm for the Insight laser source, both operating in the 1550 nm central wavelength regime) the akinetic laser exhibited a very flat RIN variation, which can be explained by the absence of mechanical tuning mechanisms in the laser and regarded as an advantage of akinetic all-semiconductor laser technology. On the contrary, it might be expected that swept laser source technologies based on mechanical tuning mechanisms and passive gain profile, i.e. where power varies with the gain of the material, would show a modulation in the output power at frequencies multiples of the mechanism repetition frequency. This behavior has certainly been shown in [17] and might be expected also for MEMs tunable, MEMS VCSEL, and/or spinning mirror based laser technologies.

 

Fig. 10 Estimation of percentile variations of (a) sliding RIN and (b) ortho RIN of Insight 1550 nm, 40 nm sweep range laser source.

Download Full Size | PDF

2.6 Considerations on integrating the Insight laser source into OCT imaging engines

The all-semiconductor akinetic laser technology impacts most aspects of the performance and design of a SS-OCT system, including the way the optical signal generated by the interferometer must be conveniently acquired, and how the subsequent digitized data is treated to obtain meaningful OCT information. In this section we describe the aspects of data acquisition and conditioning as imposed by the Insight laser source. The optical frequency versus time relation in mechanically tuned swept lasers often lack sufficient linearity to enable a direct conversion of the acquired interferometric signal (raw data) into an OCT image. To compensate this disadvantage, most SS-OCT systems require an external optical k-clock for variable-rate sampling [18,19] or the acquisition of a similar k-clock signal that can be analyzed to properly resample the even-in-time acquired raw data [20,21]. The akinetic all-semiconductor laser is built to inherently generate linear sweeps, removing the requirement for an external optical k-clock. It enables data that is linearly spaced in optical frequency (k-space) to be sampled at a constant sampling rate by the digitizer. This means that the acquisition of a reference rescaling signal to compensate for laser sweep non-linearity is no longer necessary, as it is no longer necessary to resample the acquired interferometric signal for k-space linearization, significantly reducing raw data-to-image computing time. No external k-clock was needed in our SS-OCT system, since this was provided from a digital sample clock output signal from the laser, with a sweep-to-sweep linearity deviation <0.002%, resulting in sampling triggers accurate within 300 fs – 10 × better than any optical k-clock. The digital sample clock provided by the laser source ensured optimal synchronization between the laser’s optical state changes, occurring each 2.5 ns (valid and invalid sweep points, see section 2.1), and data acquisition (sampling rate) during one sweep. At each new laser optical state, the clock must correspond to the acquired sample from the digitizer. During the full completion of one sweep/data acquisition, this synchronized one-to-one relation between optical states and acquired samples must be maintained, yielding acquisition of samples at constant rate of both valid and invalid optical laser states (see section 2.1). The effective sampling rate at which the digitizer must be set is imposed by the laser; it must match the laser master clock frequency, which was, in our experiments, 400 MS/s (or 2.5 ns sampling interval). This constraint requires that the digitizer must be driven by an external sampling clock signal – the Sample Clock signal supplied by the laser: Once acquired, and before proceeding with conventional data processing (e.g. fixed pattern noise removal, windowing, FFT, etc.), the sampled data from each sweep, i.e. each A-scan, must be decimated at specific samples’ index positions in order to remove those samples associated with laser transitions (invalid points) occurring during the sweep. To illustrate this concept, Fig. 11 reports recordings of power spectra output traces with no active sweep (Figs. 11(a)-11(b)) and with the laser sweeping at very low sweep rate (~8 kHz) at 1 mW (Figs. 11(c)-11(d) and 2 mW (Figs. 11(e)-11(f)) output power; and recordings of interferometric signal (Figs. 11(g)-11(h)) using a mirror as a sample in the sample arm (2 mW laser output power, interferometric signal power attenuated before the photodetector to avoid saturation).

 

Fig. 11 Effect of the decimation process on acquired (digitized) raw data to remove the invalid points. The inset in (a) represents the scales of the horizontal and vertical axes for each plot in the figure. Plots (a-f) show recorded power vs. sample traces before (left column) and after (right column) the removal of the invalid points. The insets in (c) show magnified portions of the trace. The red arrows in the 40 points (pts.) inset (leftmost) indicatively show the beginning (left arrow) and the end (right arrow) of the transition interval between two valid points subsets. The blue arrow in the 5000 points inset (rightmost) in (c) and its analogous in the 5000 points inset in (d) for the equivalent magnified portion of the traces, illustrate the effect of an incorrect reconstruction of the data set with invalid points removed. The 200 points insets in (g) and (h) illustrate the effect of the removal of invalid points on interferometric fringes signal.

Download Full Size | PDF

The reported traces show the signals before (Figs. 11(a), 11(c), 11(e) and 11(g)) and after (Figs. 11(b), 11(d), 11(f) and 11(h)) the removal of the invalid points. Figures 11(a) and 11(b) show recorded traces of the laser’s output power when the laser is not sweeping. Obviously, in this case no effect of the transition intervals is noticeable. From the plots 11(c), 11(e) and 11(f) it is possible to recognize the effects of the transition intervals/introduction of the unwanted points, recognizable as “spikes” interspersed in the trace. The 5000 points inset in Fig. 11(c) shows a magnification of the trace including two subsequent spikes (transition intervals) in the data sequence. The portion of the trace around the leftmost spike is further magnified (40 points leftmost inset in 11(c)), highlighting the beginning and the end (left and right red arrow in the inset respectively) of the transition interval between two valid points subsets. The blue arrow pointing the rightmost spike shown in the 5000 points inset in 11(c) and its analogous in 11(d) for equivalent portions of the trace observed before and after the removal of invalid points are highlighting the effect of incorrect reconstruction (stitching) of two subsequent valid points sub-intervals. The “stitching quality” of consecutive valid data sequences is determined by the implemented algorithms in the laser; we observed that sometimes (depending on the calibration process outcome) for the larger spikes, the stitching introduced small irregularities or discontinuities in the decimated signal, as pointed by the blue arrow in the inset of Fig. 11(d). We observed, during the course of the experiments, that these anomalies were limited or totally suppressed by improved versions of the laser firmware. The information concerning which samples have to be discarded is supplied by the laser; it is generated during some laser calibration processes that must be performed before enabling the sweep – e.g. a new laser calibration is required any time the user selects a different number of valid samples per sweep (sweep rate change) – and stored as an array of indexes, or, alternatively, it is available in the form of a TTL signal output from the laser (Data Valid signal; e.g. high – valid, low – discard) synchronized with the sweep and accessible in real time during each sweep. Stable and repeatable laser sweeps permit the use of the same discarding information for data decimation as generated by the recalibration process (stored array of indexes), applicable to the acquired raw data until the next laser recalibration occurs, without the need of an update at each sweep. Similar to other swept laser technologies, the Insight swept-laser technology imposes on the acquired signal an intermediate step – decimation for all-semiconductor technology versus rescaling or k-clock implementations for the others – before the application of conventional OCT signal processing techniques. The significant advantage offered by all-semiconductor technology is that decimation computing time is negligible when compared to rescaling/interpolation computing time required with other sweep source laser technologies applied to equal optical system complexity. Alternatively, linear and repeatable sweeps combined with decimation avoid the need of complex and potentially unstable implementation of k-clock schematics. In addition, the akinetic laser supplies a start sweep signal (A-line trigger) for proper A-scan acquisition, which eases the integration and synchronization of the laser sweep with scanning units.

3. Results

To validate laser performance, measurement at several sweep rates and at different coherence gate displacements were performed using three different Insight swept laser sources whose main specifications are described in Table 1. We applied the Insight 1550 nm, 40 nm sweep range laser source, first available in our labs, to investigate laser performances and imaging limits at different coherence gate displacements. For each selected sweep rate, the sample in the sample arm (SA) was fixed and positioned in the SA focal plane while varying the distance in the reference arm (RA). Data sets for 2D and 3D sample reconstructions were acquired at each coherence gate displacement, starting close to the zero delay up to the largest displacement allowed by the selected laser sweep rate and digitizer sampling rate (this latter was always set to 400 MS/s, as required by the laser source). Selecting a specific laser sweep rate means selecting a corresponding specific number of (valid) points per sweep (A-scan); combined with a fixed sampling rate, this also defines the upper limit of the coherence gate displacement for OCT imaging. The lower the sweep rate, the higher the number of points per sweep, which produces a higher limit of the coherence gate displacement from the zero point of the delay line. The sample used in the experiments was an ex vivo tooth; sequences of 2D tomograms acquired always from the same location on the sample for all the different configurations were processed and the results are presented in Fig. 12. Three different sweep rates were chosen for the measurements: 20 kHz, 100 kHz and 200 kHz, requiring 19000, 3500 and 1550 (valid) points per sweep respectively. The higher the number of points per sweep, the larger coherence gate displacement in RA it was possible to achieve before encountering the limits imposed by the system: first, the maximum mechanical displacement in the RA of the interferometer; second, the Nyquist limit (200 MHz) to the maximum analog signal that may be measured by our data acquisition system, clocked by the laser’s 400 MHz user sample clock. For a 20 kHz sweep, the RA displacement limited the maximum coherence gate. By contrast, at 200 kHz sweep rates the Nyquist limit of 200MHz signal frequencies is reached at coherence gate displacements well below the constraints of our particular reference arm mechanism. Fringe degradation could be observed from non-optimal laser calibration, i.e. non-optimal stitching of consecutive valid points in sweep sub-intervals, which might introduce unwanted frequency components in the interferometric signal falling within the selected imaging range.

 

Fig. 12 OCT imaging using the Insight 1550 nm, 40 nm sweeping range laser source, at different sweep rates and different displacements of the coherence gate. The tomograms show cross-sectional reconstruction of an ex vivo tooth. Data were acquired always on the same location on the sample for all the shown data sets. Each figure was averaged from 32 consecutives B-scans. Image size is 6 × 2.9 mm² (width × depth, in tissue), corresponding to 1024 × 100 pixels (hor. × vert.) for the 20 kHz images (topmost row) and 2048 × 100 pixels for the remaining images. Per each selected laser sweep rate, imaging range spans from zero delay (z.d.) up to the largest coherence gate displacement allowed by the system. Incident power on the sample ~2 mW. Refractive index n = 1.44.

Download Full Size | PDF

Jitter noise also induced a rise in the noise floor of the OCT signal, producing a loss in image sensitivity at larger RA optical path distances. Occasionally, fringe corruption caused the appearance of PSF side lobes, which translated into ghosted replicas of the sample along the imaging range as selected by the coherence gate position in the final OCT tomograms.

Keeping the coherence gate displacement within the useful limits, the obtained images were clear and undistorted. Undesired image disturbances introduced by the calibration process were removed by repeating the calibration. At each selected sweep rate–more evident in the 20 kHz case of Fig. 12 – it can be noticed a drop in intensity due to a natural decay of the probing radiation intensity and also due to the rise in the noise floor that limited image dynamic range (see the background outside the tooth boundary). From Fig. 12 it is also possible to appreciate how an increase of the laser sweep rate does not significantly degrade OCT imaging performance for a fixed coherence gate displacement position. Main factors that limited imaging at longer coherence gate displacements were the available number of samples for FFT computation and digitizer sampling rate (sample clock of 400 MHz, fixed), both imposed by the laser source. Figure 13 shows a comparison of single vs. averaged tomograms of in vivo skin acquired with 1550 nm central wavelength, 40 nm sweep range laser at 20 kHz, 100 kHz and 200 kHz sweep rates. The sample was positioned in the sample arm focal plane and the coherence gate positioned close to the zero delay. M-series of 32 B-scans were acquired and processed. Each frame was acquired every ~50 µs (19424 × 1024 raw data pixels, vert. × hor., including invalid points), ~10 µs (3924 × 1024 raw data pixels, vert. × hor., including invalid points) and ~5 µs (1974 × 1024 raw data pixels, vert. × hor., including invalid points) for the 20 kHz, 100 kHz and 200 kHz sweep rates respectively. No motion correction was applied to the averaged tomograms. The left column (Figs. 13(a), 13(b) and 13(c)) shows single frame results obtained at different sweep rates; the right column (Figs. 13(d), 13(e) and 13(f)) shows the respective sweep rate data set averaged 32 times. Incident power on the sample was ~2 mW.

 

Fig. 13 Single vs. averaged in vivo skin imaging at 1550 nm, 40 nm sweep range. a-c) single frame; d-f) averaged frame from 32 consecutive B-scans (M-series). Image size is 6 × 2.9 mm² (width × depth, in tissue, n = 1.44), corresponding to 1024 × 100 pixels (hor. × vert.) for a and d (i.e. 20 kHz sweep) and 2048 × 100 pixels for the others. Coherence gate location close to zero delay. Scale bar = 1 mm. Incident power on sample ~2 mW.

Download Full Size | PDF

In all three shown cases it was possible to obtain images with very similar quality and no noticeable intensity drop with increased sweep rate. Figure 14 demonstrates the capability of the system to perform very fast in vivo acquisitions of large 3D data sets and to generate 3D reconstruction of the sample. In the reported example, results from in vivo skin measurements are shown, acquired with the 1550 nm central wavelength, 40 nm sweep range swept laser source, sweeping at 109 kHz. The 3D volume data set of 1.5 GB was acquired in ~2.4 s with a 12-bit digitizer (Alazartech, mod. ATS9350). Figures 14(a) and 14(b) show the 3D reconstruction of the sample in its entirety and with a portion of the data removed to reveal the internal structure, respectively. Figure 14(c) shows a single tomogram extracted from the 3D volume corresponding to the vertical frontal tomogram as shown in Fig. 14(b) and delimited by the blue rectangle. Figure 14(d) illustrates an en face view of the sample, extracted from the 3D data set corresponding to the horizontal plane as shown in Fig. 14(b) and delimited by the green rectangle. The larger speckle patterns which can be noticed from the presented results are a consequence of the applied 3D rendering to the data. The latest generation of the Insight laser source at 1550 nm offers extended spectral bandwidth, up to 80 nm (sweep range @ 0 dB drop) in our tests, higher number of sampling points per unit depth and higher output power, up to 13.5 mW.

 

Fig. 14 In vivo skin imaging at 1550 nm, 40 nm SR with ~20 µm isotropic resolution. a) 3D sample reconstruction; b) internal view of the structure; c) cross-sectional view; overlapped rectangle correspond to overlapped rectangle (vertical plane, blue) in b; d) en-face view; overlapped rectangle corresponds to overlapped rectangle (horizontal plane, green) in b. Volume size is 5 × 5 × 2 mm³ (width × height × depth, in tissue, n = 1.44), corresponding to 1024 × 256 × 180 pixels. Scale bars correspond to 0.50 mm. Incident power on sample ~2 mW.

Download Full Size | PDF

Figure 15 demonstrates enhanced performance of the broad bandwidth light source, by comparing results obtained with the two 1550 nm laser sources (40 nm vs. 80 nm sweep ranges), both sweeping at 40 kHz, the highest achievable sweep rate with the broader sweep range Insight laser source at the time of the experiments. Figure 15(b) reveals sharper and more detailed morphological structure, not always evident in the correspondent image (Fig. 15(a)) acquired with the smaller bandwidth laser source. Figs. 16 and 17 introduce results obtained with the Insight 1310 nm laser source from ex vivo tooth and in vivo skin recorded at 100 kHz sweep rate. In all recordings, the samples were positioned close to the zero delay line. Figures 16(a) and 16(b) illustrate single frame vs. averaged frame 2D tomography of the tooth; Fig. 16(c) illustrates the en face projection of a 3D data set of the same sample. Figures 16(d) and 16(e) illustrate single frame vs. averaged frame 2D tomography of the skin. Figure 16(f) illustrates the en face projection of a 3D data set of the same sample.

 

Fig. 15 Extended sweep range of the 1550 nm laser improves OCT axial resolution (Δz, in tissue) imaging. a), b) cross-sectional tomographies (average from 32 B-scans) of ex vivo tooth; images size is 7 × 2.9 mm² (width × depth, in tissue, n = 1.44), corresponding to 2048 × 100 pixels and 2048 × 300 pixels (hor. × vert.) respectively. c), d) 1.5 × magnification of the rectangle areas in 12a and 12b respectively. Coherence gate close to zero delay. Scale bar = 1 mm. Incident power on sample ~2 mW.

Download Full Size | PDF

 

Fig. 16 OCT imaging using the Insight 1310 nm, 30 nm sweep range laser source. a), b) single frame and 32 frames average of ex vivo tooth; image size is 6 × 2.9 mm² (width × depth, in tissue). c) en face projection of ex vivo tooth 3D data set; image size is 6 × 8 mm² (width × height). d), e) single frame and 32 frames average of in vivo skin; image size is 6 × 2.9 mm² (width × depth, in tissue). f) en face projection of in vivo skin 3D data set; image size is 6 × 6 mm² (width × height). Incident power on sample was 0.5 mW. Refractive index n = 1.44.

Download Full Size | PDF

 

Fig. 17 OCT imaging using the Insight 1310 nm, 30 nm sweep range source. a) 3D reconstruction of ex vivo tooth; vol. size is 6 × 8 × 3 mm³ (width × height × depth, in tissue); b) internal view of the structure; c) cross-sectional tomography extracted from 3D data set (blue rectangle in b); d) en-face view extracted from 3D data set (green rectangle in b). Incident power on sample was 0.5 mW. Coherence gate close to zero delay. Refractive index n = 1.44. Scale bar = 1 mm.

Download Full Size | PDF

Figure 17(a) shows a 3D reconstruction the ex vivo tooth; Fig. 17(b) shows the same data set of Fig. 17(a) with a portion of the data removed to reveal the internal structure of the sample. Figure 17(c) shows a 2D tomography extracted from the 3D volume corresponding to the blue vertical rectangle in Fig. 17(b). Figure 17(d) shows an en face view of the 3D data set corresponding to the green horizontal rectangle in Fig. 17(b).

Despite the lower power incident on the sample (~0.5 mW), image quality was comparable to images obtained with the 1550 nm sources (e.g. Fig. 16(b) vs. Fig. 12, 2nd row or Fig. 16(e) vs. Fig. 13(e)).

4. Conclusions

We demonstrated, for the first time to the best of our knowledge, the performances of new all-semiconductor, akinetic swept laser sources for SS-OCT at 1550 nm and 1310 nm with no need of k-clock implementations. Results from ex-vivo and in vivo samples were presented. The employed systems performed imaging at zero delay as well as at very large (>170 mm) coherence gate displacement while maintaining sufficient image dynamic range. Auto- (laser) and cross- (system) correlation phase linearity and repeatability analysis per sweep was performed at different sweep rates. Obtained results show good laser phase linearity and repeatability (<2 mrad or <160 pm displacement sensitivity in both cases) with the laser operating close to the shot noise limit for phase-sensitive detection. Relative intensity noise was also investigated, showing a RIN percentile variation <0.2% per sweep, about 5 times smaller when compared to other [17] mechanically-tuned swept laser sources. During our experiments, robustness and wide customizability was experienced, allowing the user to select and/or adjust all the most important parameters, including laser output power, spectral bandwidth and sweep rate. The laser directly supplies a user sample clock (or electronic k-clock) signal and a start sweep (A-line trigger) signal for convenient integration of the laser source into the SS-OCT system and optimal synchronization with data acquisition hardware. One of the most attractive characteristics of the akinetic all-semiconductor laser technology is that laser performances and functionalities are defined and controlled by the laser source firmware, which can be updated as new releases became available. For example, at the time of the writing of this manuscript, a new release of the firmware was available and installed into the 1550 nm, 40 nm sweep range Insight laser source. This recent firmware release included several improvements and added functionalities, including e.g. the ability for users to specify sweeps not only by the number of (valid) points, but also by sweep rate or by optical frequency step size, offering to the user more significant or familiar set of parameters for defining the sweep characteristics. This release also included improvements in the calibration algorithms, which e.g. produced a significant reduction of power fluctuations during sweeps, with corresponding reduction of RIN.