A system for high-resolution depth-resolved optical imaging of fluorescence and absorption contrast

Abstract

Laminar optical tomography (LOT) is a new three-dimensional in vivo functional optical imaging technique. Adopting a microscopy-based setup and diffuse optical tomography (DOT) imaging principles, LOT can perform both absorption- and fluorescence-contrast imaging with higher resolution (100–200 μm) than DOT and deeper penetration (2–3 mm) than laser scanning microscopy. These features, as well as a large field of view and acquisition speeds up to 100 frames per second, make LOT suitable for depth-resolved imaging of stratified tissues such as retina, skin, endothelial tissues and the cortex of the brain. In this paper, we provide a detailed description of a new LOT system design capable of imaging both absorption and fluorescence contrast, and present characterization of its performance using phantom studies.

INTRODUCTION

Optical imaging offers exquisite in vivo functional contrast. Intrinsic and exogenous absorption and fluorescence can provide sensitivity to a broad range of in vivo parameters, beyond the scope of any other medical imaging modality.^{1, 2, 3, 4} However, the effects of absorption and scattering present significant obstacles to harnessing optical contrast in anything but very superficial tissues. In order to acquire useful in vivo measurements of intact tissue, it is necessary to develop high-speed, high-resolution imaging techniques that can push the limits of penetration depth, while retaining as much sensitivity to optical contrast as possible. Laminar optical tomography (LOT) has been developed to address this need and is able to provide high resolution, noncontact imaging of tissues to depths of up to 2–3 mm. While its spatial resolution cannot compete with that of confocal⁵ or two-photon microscopy,⁶ LOT is sensitive to both absorption and fluorescence contrast and is sensitive well beyond the scattering limit of tissue (300–600 μm).^{2, 3, 4} Optical coherence tomography (OCT) can provide high resolution imaging of tissues to depths of ∼2 mm,⁷ but OCT suffers from poor sensitivity to absorption contrast and cannot measure fluorescence.

The basic principles of LOT and its first demonstration were previously presented,^{8, 9} and to date LOT has been shown to be effective for several in vivo imaging applications. In the brain, LOT enabled imaging through the thickness of the exposed cortex in small mammals without disturbing the delicate interconnectivity of neurons and vasculature.¹⁰ For imaging of skin cancer lesions, LOT offers the potential to investigate subsurface absorbing and fluorescent structures, as well as measures of metabolism and oxygenation that may denote abnormal changes in cells below or at the margins of tumors.¹¹ In the heart, high-speed LOT has been shown to be capable of resolving transmural propagation of electrical waves using fluorescent voltage sensitive dyes.^{12, 13}

In this paper, we present a second generation LOT system and provide a detailed description of its design and performance characteristics. The new system incorporates simultaneous imaging of both absorption and fluorescence contrast, is faster and more sensitive than our first generation system, and utilizes three excitation wavelengths, making it a versatile tool for both research and clinical studies. Section 2 describes the instrument hardware and overall system configuration. In Sec. 3 special features of the system’s optical and electronic components are discussed. Calibration is discussed in Sec. 4 and measurement sensitivity is quantitatively evaluated in Sec. 5.

LOT INSTRUMENT AND SYSTEM CONFIGURATION

Basic principles

Light entering at a discrete position on the surface of scattering tissue will scatter within the tissue, encountering absorbers and∕or fluorophores along its path. Some of this light will emerge from the same surface of the tissue, at various distances away from the illumination position. LOT’s ability to perform depth-resolved imaging relies upon the statistical probability that light emerging at greater distances from the illumination point has, on average, scattered more deeply into the tissue. This is illustrated schematically in Fig. 1a. By measuring the emerging light for a range of source positions, and source-detector distances, it is possible to resolve and map subsurface structures to depths of approximately the same order as the source-detector separation.

Figure 1.

Principles of LOT. (a) Scattered light detected at distances further from the illumination position has, on average, traveled more deeply into tissue. (b) LOT acquires raster-scanned measurements, with detectors offset from the source position.

LOT’s measurement geometry is similar to that used in diffuse optical topography (DOT). However for DOT, source-detector separations are typically 2–3 cm, and light is generally delivered to and from tissue using a grid of optical fibers.¹⁴ In DOT, each “source fiber” is coupled to an individual light source (such as a laser diode), or to a fiber-switch, which directs source light into different fibers sequentially. Each DOT “detector fiber” is typically connected to an individual detector. This approach can lead to significant multiplexing issues, including the need for variable detector gain normalization for different source positions, and frame rate limitations.¹⁵ DOT depth-sensitivity is generally several centimeters, with resolution between 5–10 mm.

For LOT, we wish to use very small source-detector separations between 200 μm and 3 mm, thereby providing sensitivity to contrast within the top few millimeters of tissue and achieving much higher resolution than DOT. While such measurements could feasibly be achieved using a dense fiber-optic array in direct contact with the tissue, LOT has been implemented using a free-space “flying spot” approach. The benefits of this implementation are significant. Only a single laser source is required, and source positions can be rapidly changed via galvanometric scanning of the focused beam. This allows very high frame rates (up to 100 fps), acquisition of dense (or variable) source-position grids (e.g., 200×200 points), and widely adjustable fields of view (e.g., from 0.5 mm to 3 cm). This novel detection geometry requires only as many detectors as the desired number of source-detector separations [e.g., d₁−d₇ as shown in Fig. 1b]. As the source position varies, so too does the effective position of each detector such that the relative separation between each detector and the source remains constant. This means that each detector’s gain can be tuned to match its typical signal intensity for its particular source-detector separation and that its gain does not need to be modulated during acquisition. The benefits of this approach over fiber-optic grid-based measurements in terms of cost, speed, and dynamic range are therefore significant.^{16, 17, 18, 19, 20}

LOT instrumentation most closely resembles a laser scanning confocal microscope. However, rather than blocking off-axis descanned light with a pinhole, LOT measures this descanned scattered light with a linear array of detectors. The wider the separation between the focus of the incident light and the lateral position at which scattered light is detected, the deeper on average the light has traveled into the tissue. The three-dimensional (3D) distribution of optical contrast represented by these measurements can be deduced using models of light propagation, and a tomographic style image reconstruction.^{8, 9, 21} LOT is therefore a hybrid between laser scanning microscopy and DOT providing rapid, noncontact, flying-spot data acquisition, with higher resolution absorption and fluorescence imaging performance than DOT.

While others have implemented laser-based flying-spot techniques for acquisition of optical signals from tissue, to our knowledge none have done so in the same way as LOT. Ramanujam et al.²² used a steered laser beam to illuminate tissue but collected nondescanned light coming back from the tissue and therefore could not discriminate signals from different depths. Kepshire et al.²³ used a scanning laser but acquire remitted light using a wide-field camera. This allowed signal emerging at different distances from the illuminating beam to be segmented from the camera image and used for depth discrimination, but the system’s speed and dynamic range is limited by the camera’s frame rate and saturation effects compared to using a descanned detector array as used in LOT. Another implementation of LOT-like measurements uses lines of different spatial frequencies projected onto the tissue’s surface and imaging of remitted light using a wide-field camera.²⁴ This approach is the spatial frequency-domain equivalent to LOT’s scanning of a focused laser spot (delta function). Measurements with higher spatial frequency illumination yield more superficially weighted signals, whereas lower spatial frequencies yield deeper sensitivities. While having the benefits of simplicity and noncontact measurements, this method also faces multiplexing and dynamic range limitations based on the properties of camera imaging compared to descanned detection.¹³

Optics hardware

The second-generation LOT system described herein was constructed using standard optoelectronic components but was optimized to be compact and portable with a small footprint to facilitate laboratory and clinical use. The major components and layout of the system are shown in Fig. 2. The laser board [Fig. 2a] includes three air-cooled lasers (488 nm model 85-BCD-030–115, 532 nm model 85-GCA-020, and 638 nm model 56RCS004∕HS all from Melles Griot), which output linearly polarized beams at 30, 20, and 28 mW, respectively. These powers differ from the power used for imaging as the power is diminished through the system’s optical components. For clinical imaging, power is deliberately attenuated to a total of 3 mW. Digital shutters (Uniblitz LS3T2 driven by VMM-D4) are positioned in front of each laser to control the illumination wavelength and duration. The laser beams are then collinearly aligned and focused into a 3 μm single mode polarization maintaining optical fiber (PM460-HP, Thorlabs). The optical fiber is used to deliver the linearly polarized light to the upper optical board housing the scanning optics and detectors. Figure 2b shows the layout of the upper optical board. The fiber output end is rotated to produce horizontal, p-polarized light, which is used for illumination in the system. The p-polarized light emerging from the delivery fiber is collimated, passed through the first polarizer, and then sent into the polarizing beam splitter cube (PBS-03 PBB 007 Melles Griot), such that any residual s-polarized light (vertically aligned) is rejected. The collimated beam emerging from the PBS passes through a triple transmission-band dichroic filter (Di01-T488∕532∕638, Semrock) and is incident upon the x-mirror of a set of galvanometers (6215HB with 6 mm mirrors and two 671215H controllers, Cambridge Technology). The beam reflects off the orthogonal y-scanning mirror and through a scan lens (currently a 30 mm focal length, 1^″ diameter achromatic doublet, AC254–030-A1, Thorlabs). This lens focuses the light at an intermediate focal plane 30 mm in front of the scan lens. As the galvanometer mirrors are steered, they change the angle of the collimated light causing the focused spot to move in a bidirectional pattern in this image plane. This plane is then imaged onto the sample’s surface by an objective lens (currently a 75 mm focal length, 1^″ diameter achromatic doublet, AC254–075-A1, Thorlabs). The optical path before the objective is bent by 90° using a mirror, to allow the main optical setup to be mounted on a vertically translatable horizontal platform. This allows the beam’s focus to be adjusted without the need to elevate the sample. The 45° mirror can also be replaced with a dichroic filter to allow simultaneous imaging of the same optical field (at a compatible wavelength) using a CCD camera, e.g., for two-dimensional multispectral imaging or for speckle flow imaging.²⁵

Figure 2.

Optical layout of LOT. (a) Layout of “laser board” combining three lasers into a single mode fiber prior to launch into (b) optical layout of combined absorption and fluorescence LOT. On-axis light is indicated by solid lines and one set of off-axis light is indicated by dashed lines.

The backscattered light from the sample passes back through the objective, through the scan lens and onto the galvanometers where it is descanned. For absorption imaging, the light reflected from the galvanometer mirrors with the same wavelength as the illuminating light passes through the dichroic filter and into the polarizing beam splitter, where the s-polarized component is reflected at 90°. The returning light is composed of specularly reflected light from the sample and from intermediate optics, and scattered light emerging from the sample. The former will maintain its p-polarization, whereas the latter is a mixture of s- and p-polarizations. The polarizing beam splitter reflects only the s-polarized light, helping to remove the effects of specular reflections. While this discards around 50% of the scattered light signal, this is more efficient than using a 50:50 beam splitter, which would provide only 25% of the available signal and would not reduce the detection of specular reflections.

The light reflected by the PBS passes through a second polarizer to remove any residual p-polarized light and through a 150 mm focal length lens (2^″ diameter achromatic doublet, AC508–150-A1, Thorlabs). This lens produces an image of the scanning spot at its focal plane. Since the light has been descanned, the image of the spot remains stationary in this plane, even as the galvanometer mirrors translate the position of illumination beam on the sample. A 16 element linear photomultiplier tube (PMT) array (R5900U-01-L16, Hamamatsu) positioned in this plane captures light at different radial positions relative to the center of the image of the spot. A 1 mm tall slit in front of the PMT limits the vertical extent of the detection area, so the elements of the detector array distinguish peripheral light that has scattered more deeply into the tissue from the superficially scattered light at the spot’s center. Any 7 of the 16 channels can be selected for a given experiment, allowing a variety of effective source-detector separation configurations to be achieved. We chose to acquire only seven channels per array based on previous studies that suggested more detector positions would result in data redundancy.⁸ Choosing 7 separations from 16 options provides flexibility while also limiting the number of required amplifier and detection channels.

In our first generation LOT system⁸ a linear fiber array, composed of tightly packed 200 μm fibers, was placed in the image plane. Each fiber delivered light to an avalanche photodiode detector. The system reported here improves upon this design by omitting the linear fiber array and directly focusing the image plane onto a PMT array. The PMT improves the bandwidth and noise equivalent power of the system. One disadvantage of the system’s PMT array is the 1 mm pitch (0.8 mm active area) of each detector element. The magnification of the avalanche photodiode setup was 1×, such that the 200 μm pitch between fibers composing the detector array corresponded directly to a 200 μm source-detector separation at the tissue. The larger pitch of the PMT detectors requires the inclusion of a 5× magnification into the detection optics (although the ability to adjust this magnification is quite valuable for varying effective source-detector separations).

For fluorescence imaging, emission light from a fluorophore in the sample will follow the same path back through the system but will be reflected 90° by the triple-transmission dichroic. A long-pass or band-pass optical filter (depending on the type of fluorophore) is used to further block the excitation light. As with the absorption detection path, the fluorescence path also contains a 150 mm focal length lens (2^″ diameter achromatic doublet, AC508–150-A1, Thorlabs), a secondary polarizer, a slit, and a linear array PMT detector.

Figure 3a shows a photograph of our second generation LOT system. A computer and all power supplies sit on the bottom layer, the laser board is enclosed on the middle layer, and the scanning and detecting systems sit on a 12^″×18^″ optical breadboard that is supported by a z-translation stage allowing the system to be easily focused. The overall footprint of the system is 18^″×24^″.

Figure 3.

Photograph and electronic layout of LOT. (a) Photograph of the LOT system. Removable boards can be positioned to fully encase optics and electronics. (b) Wiring diagram of LOT showing data flow and control signal layout.

System electronics and control

Light incident on the PMT detectors is converted into a time-varying current, which is immediately passed through a selectable-gain transimpedance amplifier (TIA, OPA380, Texas Instruments), a low pass filter (LTC1565–31, Linear Technology), and a voltage amplifier (OPA2227, Texas Instruments) as shown in Fig. 3b. Because the detected light is descanned by the galvanometers, the image of the scanning spot at the PMT array remains stationary but varies in intensity over time as the galvanometers steer the beam over the tissue’s surface. Nevertheless, the detectors further from the center of the spot will, on average, always detect lower light levels than those closer to the spot’s center. The voltage amplifier for each detector can therefore be adjusted to produce approximately equal voltage output levels when a typical object is being imaged. This provides optimal gain for each channel, reduces the interchannel cross talk and utilizes the full digitization range of the data acquisition cards.

A personal computer with a 3.0 GHz, Pentium 4 processor and 4 GB of RAM is used to acquire data and control the system. The connectivity of the system is shown schematically in Fig. 3b. The computer contains three data acquisition (DAQ) cards: two 14 bit, 8 channel, simultaneous sampling 3 MS∕s per channel analog input cards (National Instruments PCI-6133) and a 12-bit, 8 channel, 1 MS∕s analog output card (National Instruments PCI-6713). All data acquisition is synchronized using a custom designed graphical user interface written using MATLAB^™ (Mathworks).

A major challenge for controlling the system was to properly synchronize the movement of the galvanometers with the data acquisition, and to reliably reorder the data stream acquired for each frame into a square image corresponding to the bidirectional scan. This was achieved by synchronizing the DAQ cards such that each voltage output to the galvanometers, and each input read from the PMTs are driven by the same clock signal and therefore issued simultaneously. At very high scan speeds, the limited size of the analog output board’s buffer is overcome by driving analog output with a clock synchronized with PMT sampling but at a lower divisible frequency. Several data points are acquired as the galvanometers steer the beam to the next position. Using this synchronized acquisition, frames can be acquired continuously at up to 4500 lines∕s for over 20 s, limited only by the speed at which data can be streamed to disk. Image reformation is simplified because exactly the same number of data points is acquired per linescan.

Data acquisition is initiated by a digital trigger signal that is generated by the analog output card. The fine dotted line in Fig. 3b indicates the flow of the digital trigger signal. To acquire a single frame with one laser, the analog output card, upon receiving the initiation trigger, opens one of the shutters by sending it a 5 V signal. At the same time the analog input cards are initiated by the digital trigger to acquire the data from all the PMT channels as well as the “x-position” feedback signal available from the galvanometer controller boards. For imaging with multiple lasers, different laser shutters can be synchronously opened and closed between frames. This sequence can easily be changed, for example, to repeat multiple frames for the first laser before switching to another laser.

The MATLAB™ based interface program allows for control of the number of x and y pixels, pixel acquisition rate, field of view, and laser illumination sequence. Multiple frames can be acquired continuously or individually. For individual scans, the acquired data are stored in the PC computer memory, reshaped, and then immediately displayed on screen. For fast continuous scans, the acquired data are streamed directly to disk, and processed after completion of data acquisition.

SPECIAL CHARACTERISTICS OF LOT

Comparison between LOT instrumentation and a confocal microscope

The LOT system is similar in design to a confocal microscope, with several major differences. (1) There is no z-scanning of the beam’s focus. Z-scanning is not performed with LOT because depth resolved information is acquired from the off-axis backscattered light (which is blocked by a pinhole in a confocal microscope). (2) Most confocal microscopes measure fluorescence, however, LOT simultaneously measures both absorption and fluorescence. For absorption measurements, the illumination light is the same wavelength as the detected light, so backreflections within the system must be carefully controlled. (3) LOT has low magnification and uses a low numerical aperture (NA) objective lens. Confocal microscopes typically use high magnification objectives with high NA to minimize depth of field (to improve axial resolution) and to maximize light collection. LOT does not require high magnification and does not rely on depth of field for its z-resolution. In fact the lower NA objective used for LOT injects light more deeply into tissue than a high NA lens, thereby improving LOT’s achievable penetration depth. (4) LOT can scan fields of view up to 3×3 cm² compared to less than a few millimeters for confocal microscopes.

Detecting off-axis light

The distances between the optical elements in a LOT system must be carefully optimized to retain as much of the off-axis scattered light as possible. As shown in Fig. 2b, the returning off-axis light propagates at an angle with respect to the system’s optical axis. Therefore, the distance between the scan lens and the imaging lenses must be kept as short as possible to prevent the transverse displacement of these off-axis beams from becoming too large, and also to prevent the loss of higher NA remitted light. All of the optical components after the scan lens were also chosen to be as large as possible, to reduce the obstruction of off-axis light: 6 mm galvanometer mirrors, a 25.5×35.6 mm² dichroic filter, a 38.1×38.1×38.1 mm³ polarizing beam splitter, a 2^″ diameter polarizer, and 2^″ diameter imaging lenses are used. Another important alignment parameter is the distance between the galvanometers and scan lens, which must equal the focal length of the scan lens. If this is not the case, the position of the returning scattered light on the galvanometers will not be centered and will sweep back and forth on the galvanometers creating an uneven field of view.

Optical magnification (m) and source-detector separations

The objective lens of the LOT system has a magnification of 1, leaving the optical magnification of the system (m) to be determined by the ratio of the focal length of the imaging lens (f_imaging) in front of the PMT to the focal length of the scan lens (f_sc), see Fig. 2b. In the case of our system,

$$m=\frac{f_{\text{imaging}}}{f_{\text{sc}}}=\frac{150}{30}=5.$$ (1)

Theoretically, the minimum nonzero separation distance between a source and a detector channel on the object plane (r_sḏmin) is the distance between two adjacent PMT channels (r_channs, 1 mm) divided by the optical magnification of the system,

$$r_{\text{sd}̱\text{min}}=\frac{r_{\text{channs}}}{m}=\frac{1\text{mm}}{5}=200\mu \text{m}\text{.}$$ (2)

Currently, the maximum physical distance between the two furthest channels on the linear PMT array is 15 mm. Therefore, the maximum source-detector separation distance (r_sḏmax) on the object plane, for our current configuration is

$$r_{\text{sd}̱\text{max}}=\frac{r_{\text{channs}}}{m}=\frac{15\text{mm}}{5}=3\text{mm}.$$ (3)

Therefore, by changing the system magnification, the minimum and maximum source-detector separation distance on the object plane can be easily adjusted to within the limits of the size of the optical elements and achievable signal to noise.

Field of view

We refer to the field of view of the LOT system as the full imaging region on the surface of a sample that is scanned by the incident laser beam. The size of this field of view is restricted by the physical size of the scan lens, which apertures the scanning beam. The amount of linear displacement of the focused beam for a given angular deviation of the galvanometer mirrors is given by

$$d=m_{\text{obj}}{f}_{\text{sc}}\text{\hspace{0.17em}tan}\left(2\theta \right),$$ (4)

where f_sc is the focal length of the scan lens and m_obj is the magnification of the objective lens [see Fig. 4a]. For larger fields of view, the 30 mm focal length, 1^″ diameter scan lens is replaced with a 2^″ diameter, 50 mm focal length achromatic doublet (01 LAO 803, Melles Griot). This provides a bigger aperture, as well as a larger scan distance per degree of deviation of the galvanometer.

Figure 4.

Effects of galvanometer scan angle and speed. (a) Changing the angle of the collimated light entering the scan lens causes a lateral scan of the focus in the primary image plane. (b) The angular deviation of the galvanometer determines its maximum speed. The feedback voltage from the galvanometers represents the real position of the galvanometer. The galvanometers will fail at scan angles (input voltages) that exceed the limit for a given linescan speed.

When imaging dynamic changes, the size of the field of view may become limited by the galvanometers’ scanning speed. While the x-galvanometer is driven with a bidirectional triangular wave at high speeds, the mirrors are unable to reach the limits of this motion, smoothing the effective triangle wave, and reducing the system’s field of view. Ultimately, the galvanometer mirrors will fail at high speeds when they exceed an acceleration level governed by the mirror size and their ability to dissipate heat. The galvanometers provide an analog feedback signal that indicates the true position of the galvanometer mirrors. Figure 4b shows the results of driving the galvanometer mirrors, where the maximum drive-signal voltage is compared to the actual maximum feedback position voltage for a range of different line-scan speeds. For these mirrors, ±1 V of drive voltage represents ±2° of mirror motion. The plot demonstrates that for slow scans, such as 100 lines∕s, the input voltage can reach ±9 V without distortion. However, for fast scanning at 6000 lines∕s, the input voltage can only reach ±1.5 V (±3°) without significant distortion, and ultimately failure of the mirrors at above ±2.5 V (±5°). In practice we use the galvanometers to scan up to around ±3 V (±6°) at speeds of up to 4500 lines∕s, optimizing between field of view, number of pixels, and acquisition speed as discussed in Sec. 3E.

Speed limitations

To obtain optimal signal to noise, LOT system’s bandwidth is controlled to be nearly double the maximum desired pixel rate. Optimal signal gain in the TIA circuitry is obtained for smaller bandwidths, so the circuitry incorporates three switchable feedback resistances to allow adjustment of the TIA’s bandwidth from dc to 60 kHz (10 MΩ resistor) to dc to 2 MHz (50 kΩ resistor). For high-speed imaging a 1 MHz TIA bandwidth is used and a secondary dc to 650 kHz low-pass filter is used to reduce higher-frequency noise not eliminated by the TIA. Therefore the nominal pixel-rate maximum is currently around 325 kHz (although the hardware is capable of sampling at 2.5 MHz).

The scan speed of our system is therefore limited by either the speed of the galvanometer mirrors, or the maximum pixel rate. At 4500 lines∕s scanning (2 V maximum galvo input) the system can acquire a maximum of 70 points per linescan (limited by the 325 kHz pixel rate), and so can acquire 70×70 source-position data at 62 frames∕s. For 100 frames∕s imaging, the system can acquire 45×45 source-position data for the same field of view (limited by the galvanometer linescan rate). 100×100 pixel images can be acquired at 45 frames∕s (limited by the galvanometer linescan rate). These relations are given by

$$Q_{\text{pix}}=Q_{\text{linescan}}\times N_{X\text{pix}}$$ (5)

and

$$Q_{\text{frame}}=\frac{Q_{\text{linescan}}}{N_{Y\text{pix}}},$$ (6)

where Q_pix is the pixel rate, Q_linescan is the linescan rate, Q_frame is the frame rate, N_Xpix is the number of x-pixels (source-positions), and N_Ypix is the number of y-pixels.

CALIBRATION

LOT measurement calibration

Raw LOT data measurements M are affected by various systematic errors and can be modeled by the following equation:

$$M_{s,d}=g_d\left(S_{s,d}+R_{s,d}+D_d\right),$$ (7)

where S_s,d is the true data for source position s and detector d, R_s,d is the signal contributed by reflections from intermediate optics, D_d is the amount of signal contributed by the background and stray light of detector d, and g_d is the gain contributed by the detector d and its associated TIA circuit. In order to compensate for these factors, the following additional calibration scans are captured: A “no sample” scan M_s,d(ns) is performed in which the sample is removed from the scan plane resulting in a data set with information about the gain, reflections, and background. Subtracting the no sample scan from the sample scan results in the measurement,

$$M_{s,d}-M_{s,d}\left(\text{ns}\right)=g_d\left(S_{s,d}+R_{s,d}+D_d\right)-g_d\left(R_{s,d}+D_d\right)=g_d{S}_{s,d}.$$ (8)

A “no laser” scan M_s,d(nl) is performed resulting in a data set with information about the gain and background and a “light” M_s,d(l) scan is performed in which a uniformly illuminated surface is scanned without lasers resulting in a data set with information about the gain, intensity of light (L), and background. Subtracting the no laser scan from the light scan yields,

$$M_{s,d}\left(\mathrm{\text{l}}\right)-M_{s,d}\left(\text{nl}\right)=g_d\left(L+D_d\right)-g_d{D}_d=g_{d}L.$$ (9)

Dividing the results of Eq. 8 by the results of Eq. 9 provides data that represents the sample scaled by a constant,

$$\frac{M_{s,d}-M_{s,d}\left(\text{ns}\right)}{M_{s,d}\left(\mathrm{\text{l}}\right)-M_{s,d}\left(\text{nl}\right)}=\frac{S_{s,d}}{L}.$$ (10)

Figure 5 illustrates this sequence of operations performed on a LOT scan of a piece of thread embedded in homogeneous scattering background medium.

Figure 5.

Calibration of LOT data. Raw data (one separation) showing two cotton fibers of different colors at different depths (“sample”). No sample, light, and no laser scan are used to form a calibrated image. The bright dot in the no sample scan is a system reflection. The calibrated image is shown on the right.

For cases in which the percent change in signal is the desired measure (as is often appropriate for absorption imaging¹), the number of calibration scans can be reduced to only the no sample scan. The percent change is then calculated by

$$\frac{M_{s,d}-M_{s,d}\left(\text{ref}\right)}{M_{s,d}\left(\text{ref}\right)-M_{s,d}\left(\text{ns}\right)}=\frac{S_{s,d}-S_{s,d}\left(\text{ref}\right)}{S_{s,d}\left(\text{ref}\right)},$$ (11)

where M_s,d is a scan of the region of interest, M(ref) is a scan of a reference region (or the same region at a different time).

Calibration of source-detector separations

As described above, the anticipated distance between the incident and detected light at the tissue surface, for each detector channel, can be calculated based on the lenses and alignment of the system. However, in practice, the true values of the separations may deviate from the expected values, and it is critical to know the actual physical separations for quantitative analysis of LOT data.²⁶ Accurate measurement of the source-detector separation distances, as well as the system’s field of view, can be obtained by imaging a translucent ruler against a scattering background. Figure 6 shows “raw” LOT images of such a ruler. The most striking aspect of the images is the appearance of a shadow of the ruler in the wider detector channels. As the source beam scans over a dark line on the ruler, the amount of detected light decreases, creating the darker image of the ruler. However, when the imaged position of the detector coincides with a dark line on the ruler, the amount of scattered light detected also decreases. The shadow therefore corresponds to the point when the detector position is passing over the lines on the ruler and the distance between the primary lines and the shadows in each image corresponds to the effective source-detector separation in the object plane. Because the image is already of a ruler, it is simple to calculate the field of view of the system and the “pixels per millimeter” of the data based on estimates of “a” and “b.” This allows the value of each separation in millimeters to be deduced. While these empirical results usually agree well with the predicted separations based on the system design, a “ruler” image is typically acquired before or after each LOT measurement session, to ensure that the relative positions of the source and detector positions within the system are carefully recorded for each data set.

Figure 6.

LOT ruler calibration scan. Tic marks appear in the image when the source is positioned over a tic mark on the ruler while shadows form when the detector is positioned over the tic mark on the ruler. When the detector and source are aligned (measuring on axis light) no shadow appears (left). When the detector is offset from the source (middle and right images), the source-detector distance is indicated by the shadow distance in the image. The distance is determined by measuring the shadow distance in pixels (line “a”), and using the pixels to length conversion found using the number of pixels in line “b.”

SENSITIVITY CHARACTERIZATION

Characterization of LOT sensitivity to absorption and fluorescence contrast

Phantom and measurement setup

To characterize the performance of our second generation LOT system, a set of phantom tests were performed. A phantom was designed to create an accurate model of high-resolution structure with known absorption and scattering properties to facilitate comparison of the LOT data with theoretical predictions. Aqueous tissue phantoms are a common choice for DOT measurements, since scattering intralipid and absorbers such as blood can be easily mixed into agarose and formed into tissuelike blocks. However, to create fine-structure like blood vessels, tubing containing an absorbing contrast would need to be embedded into the intralipid phantom. We have thus far been unable to identify tubing that sufficiently matches the refractive index of aqueous agarose such that it does not create reflections that are difficult to account for in modeling of experimental results. It was therefore necessary to develop a phantom with inclusions providing known contrast in accurately controlled positions without introducing a refractive index mismatch from the presence of tubing.

Our chosen phantom started with an epoxy resin base (7132, Douglas and Sturgess Inc., Richmond, CA) with an absorbing dye (Pro-jet 830NP) and scattering Alec Tiranti “Superwhite” opaque white polyester pigment added to create a homogeneous background, tissue-equivalent phantom. The recipe, based on that described by Hebden,²⁷ assumed a reduced scattering coefficient of 0.8 mm⁻¹ at 800 nm for a 1 g∕kg mix of white pigment to resin. Values of anisotropy (g) and reduced scattering coefficient $\left({\mu }_s^{\prime }\right)$ at 532 nm were estimated based on the measurements in Ref. 28. The absorption properties of the dye mixture were carefully measured using the adding-doubling method prior to mixing the phantoms. Two “background phantoms” were created from this epoxy; the first of which had absorption coefficient μ_a=0.2 mm⁻¹ and ${\mu }_s^{\prime }=0.6{\mathrm{mm}}^{-1}$ and the second had μ_a=0.2 mm⁻¹ and ${\mu }_s^{\prime }=1{\mathrm{mm}}^{-1}$ (based on typical values of low and high scattering coefficients in living tissue at 532 nm). Unlike traditional epoxy phantoms, these background phantom mixtures had no epoxy hardener added, so that they remained liquid. To mimic vessels the same resin materials were used, but more absorber and epoxy hardener were added. The vessel-mimicking mixture was then injected into pe50 plastic tubing with 500 μm inner diameter and allowed to harden. After hardening, the tubing was peeled away leaving thin, pliable phantom sticks made completely of cast resin. The first stick (stick I) had μ_a=1.5 mm⁻¹ and ${\mu }_s^{\prime }=0.6{\mathrm{mm}}^{-1}$ and the second stick (stick II) had μ_a=0.6 mm⁻¹ and ${\mu }_s^{\prime }=0.6{\mathrm{mm}}^{-1}$. The benefit of this method is that the cast phantom sticks have an almost identical refractive index to the unhardened background phantom mixtures. Therefore, the sticks could easily be immersed into the liquid mixtures and accurately positioned at different depths, without the drawbacks associated with using plastic tubing.

For fluorescence measurements, aqueous solutions of Fluorescein and Rhodamine were injected into pe50 tubes, and the end of the tubes were sealed. Two concentrations, 0.8 and 0.2 mg∕ml, were adopted for both fluorescent dyes. The same resin-based background solutions were used for fluorescence tests. While the fluorescence phantoms utilize tubing, the refractive index mismatch is less in the liquid resin background than it is in the aqueous phantoms. Furthermore, fluorescence measurements are less sensitive to reflections because light emerging from the phantom at the emission wavelength of the dye must have originated from within the tubing, although smaller signals are expected than would be detected in the absence of tubing.

To acquire measurements, each vessel-like stick or fluorophore-filled tube was submerged into the homogeneous resin solution using a frame held by a micrometer positioning stage. For each depth, LOT scans with a 6.3 mm field of view were acquired over the region where the stick was submerged. For each scan, seven detector channels were simultaneously acquired for seven different source-detector separations. After acquiring the data, the position of the stick was lowered and imaging was repeated, as illustrated in Fig. 7.

Figure 7.

LOT phantom measurements. Left: absorbing sticks or fluorophore-filled tubes were lowered into the background phantom media. Top row: LOT raw data acquired using four different detectors on an absorbing stick. Bottom row: each image raw-data was averaged across rows and plotted as a function of source y position. For absorption, contrast was defined as the difference between the average minimum calibrated signal $\min \left(S_{s,d}^A\right)$ and the average background value ≈S^A(ref), with this difference then divided by S^A(ref). For our fluorescence phantom, S^F(ref)=0 and signal is positive so contrast was defined as $\max \left(S_{s,d}^F\right)$.

Analysis of phantom data

Figure 7 shows an example of the acquired raw data from four detector channels when the absorbing stick is located at a depth of 0.5 mm. To calculate the measured absorption contrast for each image, pixel values were averaged along the x-direction and then plotted against their y positions (see Fig. 7, bottom plots). Note that detector positions were oriented to be offset in the y-direction. Absorption contrast was quantified by subtracting the calibrated average minimum signal $\min \left(S_{s,d}^A\right)$ from the average background value $\approx S_{s,d}^A\left(\mathrm{ref}\right)$, and the difference was divided by the background value as given by Eq. 11.

There are many ways to analyze and interpret LOT data, ranging from simple inspection of the raw data, which provides subsurface enhancement of deeper absorbing and fluorescence objects,²⁶ to reconstructing 3D images of the depth-resolved structures.^{2, 3, 4} The latter approach is complex and can be approached several ways, each of which can provide different imaging performances.²¹ The analysis in this paper is therefore focused only to quantifying the measurement sensitivity of LOT: its ability to detect and resolve buried objects within scattering. The theoretical limits of LOT imaging performance will be the subject of a future publication.

Absorption contrast versus target depth

Figure 8 shows the measured contrast (on a log scale) as a function of target depth for stick I (μ_a=1.5 mm⁻¹) submerged in background solution I (μ_a=0.2 mm⁻¹, ${\mu }_s^{\prime }=0.6{\mathrm{mm}}^{-1}$) (a) and stick II (μ_a=0.6 mm⁻¹) submerged in background solution II (μ_a=0.2 mm⁻¹, ${\mu }_s^{\prime }=1{\mathrm{mm}}^{-1}$) (b). The green laser (532 nm), with 3.84 mW at the focus of the objective lens, was used for the measurements. Each line shows the contrast in the raw data for each detector channel: ranging from a source-detector separation of 0–2.4 mm. The 0 mm channel shows the lowest contrast for all depths, but with a stronger decay in contrast as the stick moves deeper into the scattering background. This reflects the more superficial sensitivity of the 0 mm channel, as would be expected. The wider channels exhibit higher relative contrast for deeper objects compared to narrower channels, although the contrast decreases for all channels as the object moves deeper into the scattering background. This demonstrates the sensitivity of wider channels to both superficial and deeper structures. The difference in depth sensitivity between the narrower and wider detectors is the feature that provides LOT its ability to perform depth-resolved imaging.

Figure 8.

LOT phantom imaging results. (a) Absorption contrast vs target depth for high contrast low scatter phantom and (b) low contrast high scatter phantom. The dashed lines indicate the noise floor for the three incident power levels indicated, for the r_s,d=0 mm channel. (c) Similar experiments were carried out using fluorescence contrast with low scatter and (d) high scatter. (e) Noise floor for all channels for each of the three laser intensities when imaging “background I”. (f) Noise floor plots for different scanning speeds.

The “noise floor” values are defined as the standard deviation of repeated measurements of the background signal, divided by the mean background signal. The three blue dashed lines indicate the noise floors of the r_s,d=0 mm channel with incident laser intensities of 0.35, 1.14, and 3.84 mW, respectively. These measurements demonstrate how varying the laser power can impact the ability of the system to resolve contrast at depth. They also reveal the position at which the contrast cannot be resolved from the system noise.

Comparing Figs. 8a, 8b, the maximum depth that a channel can detect depends on the separation distance between its source and detector, the noise floors (intensity of the laser), the optical properties of the background, and the contrast of the target. For the best-case: a high contrast target and low background scattering [Fig. 8a] the detection limit is greater than 3 mm deep. For the worst case: a low contrast target in a highly scattering background [Fig. 8b], the depth-detection limit is around 2.3 mm (note that the r_s,d=0 mm data value reaches its noise floor earlier since it is essentially zero, and is largely insensitive to deep objects). Decreasing the laser intensity will reduce the maximum detectable depth. Measurements were also performed for all three wavelengths for stick I in background II, and stick II in background I (not shown). In all of these cases, it was clear that lower object contrast and higher background scattering both limit the depth-sensitivity of LOT. In general, the 638 nm data provided around a 0.5–1 mm deeper depth-limit than the 532 nm laser, as would be expected from the lower scattering properties of the phantom at 638 nm, which would also be true in tissue.

Figure 8e shows the noise floor for all channels at the three incident laser intensities for background I. The error bars represent standard deviations of 20 measurements. The increase in this noise floor for wider source-detector separations is the result of the lower light intensity detected by these channels (since their light has scattered further into the tissue). Figure 8f shows the effect of scanning speed (and therefore data acquisition speed) on the system’s noise floor. The scanning speed is only weakly correlated with the system’s noise floor except when scanning speeds reach 5000 lines∕s and over.

Fluorescence contrast versus target depth

To characterize fluorescence contrast, an equivalent analysis of the data acquired through the fluorescence arm of the system was performed. Because there was no fluorophore dissolved in the background solution, the background signal for fluorescence measurements was approximately zero (whereas the absorption-contrast raw data had a bright background signal with dark contrast). This means that the fractional contrast [S_s,d-S_s,d(ref)]∕S_s,d(ref) of the fluorescence data cannot be calculated as described earlier. Therefore, “fluorescence contrast” was instead defined as the maximum fluorescence signal relative to each channel’s gain $\max \left(M_{s,d}\right)∕g_d=\max \left(S_{s,d}^F\right)$ since reflections R_s,d and stray light D_d are suppressed by the fluorescence emission filter [Eq. 7]. Gain values g_d were derived from measurements made by translating a steady-state focused laser beam across the face of the linear PMT array.

Figure 8c shows measured fluorescence contrast data (on a log scale) as a function of the depth of a Rhodamine-filled tube (0.2 mg∕ml) submerged in background solution I, measured using the 532 nm laser, and a 550 long-pass filter in front of the PMT array. The fluorescent contrast decays quickly in all channels with increasing target depth. Because the fluorescence contrast is not scaled by the background signal in each channel, the absolute magnitude of the fluorescence contrast is highest in the 0 mm channel and lowest in the wider channels. However, the trends in the gradients of the contrast data are similar to those in the absorption data; the contrast in the 0 mm channel reduces at a faster rate than the contrast in the wider channels. The noise floor was measured on the Rhodamine-filled tube and divided by the channel gains to allow comparisons with the phantom data. The blue dashed lines show the noise floor for incident laser intensities of 0.35, 1.14, and 3.84 mW for the 0.6 mm source detector separation channel. Figure 8b shows equivalent data for the same tube submerged in background solution II.

For both backgrounds I and II, fluorescence contrast can be detected to deeper depths than absorption contrast, with good signal at up to 3.5 mm in depth. The reason for improved contrast of fluorescence data relative to absorption contrast is likely to be the result of the following effects. (1) Fluorescence contrast is “all or nothing” signal, where the only light that is detected at the emission wavelength must have originated within the fluorophore. In the absence of background fluorescence, this makes it easier to pick out the presence of signal from background, which can be more challenging for absorption images where the detected light will always contain signal that has only interacted with more superficial layers. This zero background $\left(S_{s,d}^F\left(\mathrm{ref}\right)=0\right)$ also means that system gains can be adjusted to amplify the contrast of the stick such that the system’s dynamic range spans from zero to $\max \left(S_{s,d}^F\right)=\Delta S^F+0$ rather than from zero to $\max \left(-S_{s,d}^A\right)=\Delta S^A+S_{s,d}^A\left(\mathrm{ref}\right)$ as required for absorption contrast. (2) Propagation of fluorescent light is slightly different to propagation of light detected in the absorption arm. This is because fluorescence emission is isotropic, whereas backscattered light carrying absorption information maintains its directionality. This can affect the relative depth-sensitivity of absorption and fluorescence measurements, even if the same source-detector separation is used. (3) The fluorescence emission wavelengths of Rhodamine are longer than 532 nm. Longer wavelength light typically experiences less absorption and scattering than lower wavelength light, which may have affected our results. Had fluorescent contrast been added to the background medium, the depth-sensitivity results would become closer to those shown for absorbing contrast, since a background signal would be present, which would decrease the useful dynamic range of the system.¹³

SUMMARY

We have described the design and construction of a new depth-resolved optical imaging tool. This second generation LOT system is capable of imaging at frame rates exceeding 100 Hz, with simultaneous sensitivity to absorption and fluorescence, and with much improved sensitivity compared to our original system. Calibration measurements were described and important aspects of the system were characterized, including fundamental attributes of LOT as well as details specific to our design. Lastly, characterization measurements demonstrating the system’s ability to detect the contrast of absorbing and fluorescent structures submerged in scattering and absorbing backgrounds of different optical properties were presented.