2.1.

FD Detector Channels

The PMT modules were used with wavelengths below 850 nm for FD measurements. The instrumentation for this portion, illustrated in the schematics shown in Fig. 1(b), is nearly identical to our previously published system,¹³ with two important upgrades, including (1) a 20 dB amplifier that was added after each PMT and (2) a programmable gain amplifier on the DAQ card. The light source consists of six intensity-modulated (100 MHz) laser diodes that cover the main spectral range of the four chromophores under study (660, 735, 785, 808, 826, and 849 nm). A bias-T combines DC bias current from an ILX Lightwave laser driver with a 100 MHz AC signal from a Marconi function generator in order to drive the laser diodes. The modulation depth of the lasers is 25 to 40%. The lasers are activated one at a time by digitally controlling a six-way RF switch, which passes the biased current to the selected laser diode. Each laser diode is mounted on a fiber launch module (Thorlabs, Princeton, New Jersey) where the output of each laser is directed into the input of a $6\times 1$ fiber optic combiner (Fiberguide, Stirling, New Jersey). The output fiber is then directed to the rotating stage, discussed in Sec. 2.3, where it acts as the source fiber in the system.

PMT (H9305-3, Hamamatsu, Japan) detectors with a 1.4 ns rise time are used to acquire the attenuated laser light that is transmitted through tissue. The spectral response range of the PMTs is 185 to 900 nm with peak sensitivity at 450 nm. The sensitivity of the PMT decreases rapidly at wavelengths longer than 850. Power to all PMTs is supplied from a single power supply. Thin film filters (Kodal Wratten Red 25, Edmund Optics) are installed on the detectors window in order to block visible light below 600 nm from entering the PMTs.

The RF output of the PMT detectors is amplified using a 20 dB RF low-noise preamplifier (Minicircuits), which also filters out the DC component of the signal. Adding this amplifier has increased the SNR by a factor of 100. The output of the preamplifier is then heterodyned with a local oscillator (LO) reference signal (100 MHz) in order to down-convert the 100.0005 MHz PMT output and produce an intermediate frequency (IF) signal (500 Hz) that can be processed more easily by the DAQ cards. The IF output of the mixer is then amplified ($100\times$) and filtered by a differential-amplifier/3-pole Butterworth low-pass filter circuit in order to reduce high-frequency noise before it is measured by the DAQ card. The resulting output signal is read by the DAQ card where it is phase-locked with a reference signal in order to obtain the phase and amplitude of the attenuated signal detected by the PMT.

The LO signal is obtained by dividing the power of a 13 dBm 100 MHz function generator output into 15 signals using a $1\times 16$ RF splitter. Each output of the RF splitter is then connected to the LO input of the mixers. One output is terminated via a 50 Ohm terminator since only 15 mixers are used. The two signal generators are synchronized and the reference signal used for the phase-lock detection is obtained by mixing a portion of the output of each signal generator. This reference signal is then connected to the DAQ card where the phase from all channels is compared with the phase of the reference signal.

2.2.

CW Detector Channels

The CW measurement system extends the range of wavelengths used to image the breast in order to better quantify the chromophore concentrations. The system consists of three laser diodes (903, 912, and 948 nm) and 15 silicon PD modules that are housed in the same enclosure as the FD components. The output of the laser diodes is intensity-modulated at 50 Hz in order to obtain better SNR through software-based lock-in detection as shown in Fig. 1(b).

The detectors used are high-precision silicon PD modules that incorporate a current-to-voltage amplifier within the module (C10439-03, Hamamatsu, Japan). These modules have an active area of $1\mathrm{c}{\mathrm{m}}^2$ and a large spectral response range (190 to 1100 nm) with peak sensitivity at 960 nm. The sensitivity of the PD modules is switchable between a low and a high range. The two PDs nearest to the laser source are set to low sensitivity in order to avoid light saturation of the modules while the remaining modules are set to high sensitivity. Thin film filters (Kodak-87 Wratten, Edmund Optics, Barrington, NJ) are installed on the detector’s window in order to block light below 800 nm. All PDs are driven by a single dual-output power supply. CW measurements only extract the change in amplitude of the light through tissue. Therefore, unlike the FD measurements, no additional components were needed to connect to the PD modules. The output of the PDs is connected directly to the DAQ cards.

2.3.

Rotary Mechanical Switch Array

A portable rack contains all the FD and CW components including laser diodes, signal generators, computer, and the detectors array. The detectors array is housed inside a custom mechanical switch as shown in Fig. 2, which was designed using SolidWorks and built in-house using ¼”-aluminum sheets. The assembly (${21}^{\prime \prime }\times {21}^{\prime \prime }$) was designed to accommodate both PD and PMT modules. The PD and PMT modules are mounted on the top circular plate, while the RF splitter and mixers are mounted on the bottom circular plate. Both circular plates rotate via a high-precision motor mounted on the bottom circular plate as shown in Fig. 2(a). This allows multiplexing of the laser source and detectors to 16 different positions, giving a total of 240 measurements ($16\text{sources}\times 15\text{detector}$ measurements) for each wavelength. The detector’s position is always fixed relative to the source position. Fiber optic cables connect the patient interface to the top square plate of the switch assembly. A small gap exists between the top square plate and the circular plate holding the detectors, which allows the circular plate to rotate. Light from the fiber optic cables is multiplexed to all detectors through holes in both the square and circular plates that align with each other as the circular plates rotate. The fibers remain fixed and thus the same fiber that was used for the source is also used for detection.

Fig. 2

Rotary switch array: Solidworks drawing of the array is shown in (a) and photos of the actual assembly are shown in (b) and (c).

2.4.

Computer and Data Acquisition

A computer running custom LabVIEW software controls the system components and enables automation of the entire system. The program allows the user to select the lasers to be used for imaging and provides advanced options that enable the user to run different customizable scans. The LabVIEW software controls the signal generators via general purpose interface bus, the rotary stage through RS232, and two peripheral component interface (PCI) DAQ cards.

A 16-bit 64-channel multifunction DAQ card (NI-PCI 6031E, National Instruments, Austin, TX) is used for acquiring the data from the PMT. The FD signals are sampled at 4 KHz and $1000\text{samples}/\text{channel}$ are acquired. The signal from each FD measurement is then phase-locked with a reference signal that is also connected to the same DAQ card. The phase and amplitude of each channel is analyzed and recorded. The CW data are sampled at 4 KHz and $2000\text{samples}/\text{channel}$ are acquired.

The multifunction DAQ card used has a programmable gain amplifier. The use of this digitally controlled amplifier ensures that the analog FD/CW signals use the maximum resolution of the analog-to-digital converter (ADC) of the DAQ card, which improves the signal’s accuracy. Using this amplifier has increased the SNR by a factor of 10. During acquisition, data from each channel is sampled for 100 ms in order to find the maximum and minimum of each channel. Once this range is found, the appropriate gain is selected for each individual channel and the signals are acquired again.

The gain of the PMT detectors is controlled by a 16-bit 16-channel static analog voltage output card (NI PCI-6703, National Instruments). The gain is set by applying voltages between 0.3 and 1.1 V in increments of 0.1 V.

2.5.

Calibration

Calibration is a critical step needed in order to get meaningful linear response of voltage data from the PD/PMT detector optical measurements. This takes into account amplitude losses and phase delays that result from differences in the detector’s response, optical fiber losses, and RF components performance. The goal of the calibration process is to produce the same response (amplitude and phase measurements) for equivalent inputs across the entire dynamic range of the detectors. Three types of calibrations were used as described below.

2.6.

Patient Interface

Figure 3 shows the schematic of two different breast interface geometries. Instead of the circular patient interface [Fig. 3(a)] that has been used with our previous generation imaging system,¹³ a pentagonal patient interface [Fig. 3(b)] was designed to accommodate various breast sizes to be imaged inside the magnetic resonance (MR) room.²⁰ One end of each of the 16 long fiber bundles is coupled into a breast MR coil and contacts the breast with limited pressure for simultaneous MR and optical imaging of patients/phantoms. The other end of each of these 16 fiber bundles is passed through a conduit in the wall and connects the source and detectors in the imaging assembly located outside the MR room. More details about this new interface are discussed elsewhere.²⁰

Fig. 3

Schematic of two different breast interface geometries: circular (a) and the new pentagonal interface (b).

2.7.

Phantom Experiments

In order to study the system in detail, tissue phantom experiments were performed on a gelatin phantom mimicking breast tissue, with a 2 cm cylindrical inclusion within, which could have varying optical properties. The bulk optical properties of the phantom were similar to those of breast tissue. It was constructed using water, type 1 agarose, 1% intralipid, and whole porcine blood at a background concentration of 15 μM. The hematocrit level of the blood was measured by a Hemocue device to be $8\mathrm{g}/\mathrm{dL}$ of the blood. The inclusion was filled with varying concentrations of blood from 0 up to 45 μM in 5 μM increments, in order to vary the absorption systematically. After each concentration, the phantom was imaged with all nine wavelengths.

Images were reconstructed using Nirfast software²¹ to determine the accuracy in recovery of the blood concentration of each run as compared to the expected concentrations. Briefly, the difference between measured data and a diffusion-based model of light propagation through the medium^22–24 is minimized to yield estimates of the optical properties of the tissue of interest. The three-dimensional reconstruction algorithm was designed to employ a priori information gained from MRI to guide the optical solution as outlined in previous work by Carpenter et al.^25,26 This technique makes the assumption that each of the segmented regions defined from the MR, adipose, and fibroglandular have similar optical properties throughout. We simplify the image reconstruction problem computationally by completely eliminating variation within regions and, thus, are able to quantify optical properties between regions but not within them.²⁶ We also use spectral priors in the reconstruction, solving directly for tissue chromophores rather than absorption (${\mu }_{\mathrm{a}}$) and scattering (${\mu }_{\text{s}}^{\prime }$) coefficients at each wavelength as outlined in previous work.^23–25,27 These methods were applied to both phantom and human subject image reconstruction.

2.8.

Healthy Subject Imaging

All human subjects imaging has been carried out under an institutional review board approved protocol, #11487, and subjects have given informed consent prior to entering the study. Imaging protocol for the human subject examination was approved by the Committee for the Protection of Human Subjects at Dartmouth-Hitchcock Medical Center. The subjects were positioned into the clinical MR breast coil and optical fibers were moved into contact with the breast tissue. If a fiber was unable to contact the breast, data from that fiber were omitted. Coregistration between optical and MR images was done using MR fiducial markers in the plane of the fibers and bilateral MR images in the axial geometry. Optical and MR data were collected concurrently, with data collections taking 12 and 30 min, respectively, and reconstructed using the methods outlined above.

3.1.

Detection Linearity/Limits

In order to determine the sensitivity and linearity of each detector, a fixed optical signal along with neutral density (ND) filters were used to look at the detector’s response as a function of incident light power. The light power was increased incrementally by removing the ND filters from the path of the laser beam while the AC amplitude and phase response of the PMTs and the AC amplitude of the PDs were recorded. The AC signal error and phase error as a function of incident power for one of the PMTs for this system is shown in Fig. 4. The results are compared to the previous generation system built in our lab¹³ in order to demonstrate the improvement of the new system over the older one. The full dynamic range of the PMTs was about $1\times {10}^5$ with the lowest signal detected at around 0.01 nW using the highest gain setting. Adding the RF amplifiers to the system has helped in increasing the SNR of the FD measurements.

Fig. 4

Standard deviation of the natural log of the amplitude (ln AC) (a) as well as the standard deviation of the phase (b) of the photomultiplier tube as a function of incident light power. The blue marks represent the results for the new system, while the red marks represent the results for the system built by McBride et al.

Figure 5 shows the response of one of the PDs as a function of light power for the two gain settings of the PD. The low range exhibits linearity over the range of 0.1 nW up to 2 uW, while the high gain setting exhibits linearity over the range of 1 pW up to 20 nW. Thus, the full dynamic range of the PDs is $2\times {10}^6$.

Fig. 5

Photodiode (PD) response as a function of incident light power for the two available gain settings of the PD.

3.2.

Gain Settings

PMT detectors nearest to the light source receive orders of magnitude more light than detectors that are farther away. In order to account for this variability, the gain of the PMT detectors must be adjusted accordingly. The gain of the PMT detectors is adjusted by adjusting an analog control voltage input, which is controlled via a static analog voltage output NI-6073 card from National Instruments.

Prior to acquisition, the PMT would be set to the lowest gain possible (0.3 V). The gain of each PMT is then increased in 0.1 V increments while monitoring the PMT output AC signal. Once the AC signal reaches a user-defined threshold, the gain of that particular PMT is set. The process continues until all PMT gain values have been set or until the gain reaches a maximum user-defined value. The system waits for 100 ms during the process of finding the proper gain before acquiring and analyzing the AC signal. However, once the gains have been set for all PMTs, the system waits for an additional 1000 ms to ensure better stability of all PMTs.

As mentioned previously, the sensitivity of the PD modules is switchable between a low and a high range using a switch on the back of the PD module. The two PDs nearest to the laser source receive much more light than the remaining modules and therefore had to be set to low-sensitivity range in order to avoid detector saturation. A high-sensitivity setting was used for the remaining modules since the amount of light received by these detectors is below the saturation point of the high-sensitivity range.

In order to improve the performance of the system even further, a programmable amplifier on board the DAQ card was used. During acquisition, data from each channel (FD and CW) is sampled for 100 ms in order to find the maximum and minimum of each channel. Once this range is found, the appropriate gain is selected for each individual channel and the signals are acquired again. This ensures that the maximum ADC resolution is used while acquiring all signals.

3.3.

Cross-Talk

There are two types of cross-talk that may affect the results of the system. The first type is an inherent property of the PMT, where increasing the light intensity of a signal while maintaining the phase influences the phase readout. This is due to space charge effects of the PMT. In order to minimize this cross-talk, we start the PMT at the lowest possible gain and increase the gain in discrete steps until the signal is above the noise floor of the system, but well below the saturation level of the PMT. Once the signal level reaches a user-defined minimum value, which is approximately the middle of the dynamic range of the PMT, the gain is set. This method ensures that the PMT anode current does not reach large values that may disturb the electric field inside the PMT and cause the phase to change.

The other type of cross-talk that may affect the measurement is interchannel cross-talk from the DAQ card. This type of cross-talk may be improved if differential measurements are used. However, due to system limitations, our system had to use the single-ended termination where all the signals share the same ground point. This causes cross-talk between channels if one channel detects a larger signal compared to its neighboring channels. However, the same method that improves the space charge cross-talk affect also improves the interchannel cross-talk. By increasing the gain of each PMT until the signal is in the middle of the dynamic range for all PMTs, the voltage read by the DAQ device is roughly the same for all channels.

3.4.

Stability and Repeatability

The repeatability of the system was determined by repeating a phantom measurement continuously while keeping all the parameters in the experiment the same. The phantom used is a silicon phantom with diameter of 86 mm. The laser light was directed to the center of the phantom. The phantom experiment was run five times and the standard deviation of the phase as well as the standard deviation of the natural log of the amplitude (In AC) was calculated. It was found that the average repeatability of the phase and In AC is 0.2 deg and 0.5%, respectively, as compared to 0.4 deg and 0.5% for our previous system.

In order to measure the stability of the system, the same repeatability experiments were performed over an entire week, ensuring that the system is turned on and off prior to the experiments. It was found that the average stability of the phase and In AC is 0.44 deg and 1.2%, respectively.

3.5.

Data Acquisition Duration and Timing

Experiment duration is very important when patients are involved in imaging. Several improvements were implemented in order to shorten the duration of the full nine-wavelength acquisition, including optimizing the way PMT gains are calculated for different source position as well as refining software procedures.

During FD acquisition, for gain routine A, the first laser is activated and the rotary stage moves until the source and detectors are aligned with the first source position [e.g., Fig. 2(a)]. The gain finding routine sets the PMTs to the proper gain and then data from all PMTs are acquired. Next, the rotary stage moves to the next source position, sets the proper gain, and acquires the data. This continues until every source position has been acquired. Once the results from the first wavelength are obtained, the system repeats the same process above for the remaining wavelengths.

Another variation to the routine above is used to reduce the acquisition time. The second routine (routine B) reduces the time required for calculating the PMT gain by adjusting how the gains are calculated. It is identical to routine A above during the acquisition of the first wavelength. However, for the remaining wavelengths that are used in the acquisition, the gain values that were found for the first wavelength are used as the initial starting values. In both gain finding routines, once the FD acquisition is finished, the system performs a similar process for the CW wavelength measurements. The only difference is that no gain finding routine is needed while acquiring data with the PD modules since PD module gains cannot be adjusted using software.

Running an acquisition with all nine wavelengths takes anywhere from 9 to 20 min, depending on what geometry and which gain finding routine are used during the acquisition. The total data acquisition time for each wavelength is shown in Table 1. In addition to the total data acquisition time at each wavelength, Table 1 shows the average time it takes for the system to determine the gain for the PMTs (column: Gain); for movement of the rotary switch array (column: Motor); and for miscellaneous steps, which include toggling the lasers, setting up the function generators, etc. (column: Misc). As can be seen from Table 1, for FD measurements, the difference in the total acquisition times is due to the geometry chosen and the gain routine used. The circular geometry setup [Fig. 3(a)] takes the least amount of time because this geometry requires only one gain set of all PMTs at a single source position. For the noncircular geometry, gain routine B approach has cut down the time of a full nine-wavelength scan from 19 to about 12 min.

Table 1

A breakout of the time spent on the main sections of the acquisition process.

3.6.

Tissue Simulating Phantom Experiments

A blood gelatin phantom was prepared as described in Sec. 2.7. The blood concentration of the phantom background was kept constant at 15 μM. The 2 cm inclusion was filled with different porcine blood concentrations starting from 0 up to 45 μM in 5 μM increments (15 μM concentration was not used). With each concentration, full FD/CW measurements were acquired and the blood concentration was recovered using Nirfast software.²¹ Figure 6 shows reconstruction results as a function of the actual blood concentration used. Hemoglobin recovery in the background was 25 μM compared to the actual 15 μM concentration. The background values were very stable across concentrations, with the norm of the residuals being 0.0054. Recovered HbT values from the inclusion showed a linear increase (norm of the $\text{residuals}=0.0066$) with an average recovery of 71% of the true blood concentration.

Fig. 6

Reconstruction values on a gelatin phantom with 2 cm inclusion. The inclusion was filled with varying concentrations of blood. MRI images of the phantom are shown with the 2 cm inclusion overlaid (a) and (b). A photo of the actual phantom is shown in (c). Reconstruction as a function of true total hemoglobin values are shown for total hemoglobin, oxygen saturation, water, lipids, scatter power, and scatter amplitude in (d). Blue circles are for the background, while red circles are for the inclusion.

3.7.

Human Subject Imaging

The combined FD/CW system and pentagonal interface has been used to successfully image nine healthy volunteers to date. A representative case is shown in Fig. 7. The volunteer was positioned such that her breasts were centered in the MR coil to minimize image artifact and then lightly compressed by the fiber optic cables to ensure good coupling. Optical data were collected in $\sim 9\min$ for FD and 3 min for CW, totaling 12 min. MRI acquisition took $\sim 30\min$, and a second optical scan was performed during the remaining time. Data from one fiber were removed due to poor coupling with the breast tissue.

Fig. 7

Representative healthy volunteer imaging by combined optical system and MRI. Different axial MRI slices are shown with overlays of total hemoglobin (a), oxygen saturation (b), water (c) and lipid (d) fraction, scatter amplitude (e), and power (f). In each case, the region with the lowest value is removed and the region with the highest value is made slightly opaque to visualize overlaid MRI. A bilateral axial image (g) shows little degradation of MRI image quality.

A three-dimensional (3-D) finite element mesh of $\sim 40\mathrm{k}$ nodes was created specific to the patient based on the MR scan and segmented into adipose and fibroglandular regions. Optical data were coregistered to the MRI using fiducial markers in line with the optical fibers and the data were calibrated absolutely versus the model. A hard priors approach was then used to reconstruct for oxygenated hemoglobin, deoxygenated hemoglobin, water and lipid fraction, scatter amplitude, and scatter power.²¹ This volunteer showed total hemoglobin concentration of 18 μM in the adipose region and 22 μM in the glandular region with high oxygen saturation (74 to 86%) as well. Water content was higher in the glandular region, while lipid content was higher in the adipose region. MRI image quality was not affected by the optical image acquisition aside from a slight distortion of one breast from the optical fibers. The volunteer reported no discomfort due to the procedure.