1. Introduction

Approximately 60,000 infants are born each year at an extremely low gestational age (ELGA) (<29 weeks). While the survival rate of ELGA infants has risen to >50% in recent years, about 25-30% of surviving infants suffer neurodevelopmental impairments, such as cognitive, sensory, and motor disabilities [1]. As a consequence of impaired cerebral autoregulation and the inability of the brain’s vasculature to withstand rapid changes in cerebral blood flow (CBF), intraventricular hemorrhaging (IVH) follows the rupturing of vessels in the germinal matrix. The severity of the IVH (grades I to IV) has a significant impact on long term neurodevelopmental outcomes. More severe grades of IVH (grade III and IV) carry a 50% risk for cerebral palsy and other intellectual disabilities [2,3]. In most cases IVH occurs within the first three postnatal days, with its pathogenesis linked to the increase, decrease and significant fluctuations in cerebral blood flow, among other risk factors [4,5]

Considering IVH is believed to be caused by poor cerebral autoregulation, monitoring CBF within the first three days of life can allow clinicians to detect physiological instabilities and make timely interventions to reduce risk of injury and improve developmental outcomes. Head ultrasound is the standard of care to detect IVH in ELGA infants [6], however it is typically used only for daily static imaging of the hemorrhage and point measurements of cerebral blood flow velocity [7], not for continuous monitoring of cerebral blood flow. This is because the need to maintain precise alignment of the transducer with the insonated blood vessel requires the use of a rigid head mount that can deform the infant skull, and the possible risks associated with tissue heating and cavitation with long ultrasound exposure, prevent the use of Transcranial Doppler ultrasound as a continuous blood flow monitor in infants [8]. Imaging modalities like positron emission tomography and arterial spin labeled magnetic resonance imaging provide only a single blood flow time point and are rarely used in this population [9,10], since the extreme premature infants are too unstable to be moved to the scanner.

Commercially available and FDA cleared cerebral oximeters, based on near-infrared spectroscopy (NIRS) optical methods, have been adopted for more than 30 years by many hospitals to noninvasively quantify hemoglobin oxygen saturation (SO₂) [11–13]. NIRS uses two or more continuous wave (CW) light sources and collects photons that pass through tissue at one or more source-detector (SD) separations to quantify attenuation changes at each respective wavelength, and under several assumptions, it estimate cerebral oxy- and deoxy- hemoglobin concentration changes and regional SO₂ [14,15]. NIRS systems operating in frequency-domain (FD) or time-domain (TD) regime, which in addition to light attenuation, also measure phase shifts or time-of-flight, are able to quantify both absorption and reduced scattering (µ_a and µ_s’) coefficients of the illuminated tissue, hence allowing the calculation of hemoglobin concentration and SO₂ with fewer assumptions relative to CW-NIRS methods. FD-NIRS and TD-NIRS, however, require more complex and expensive hardware, and generally provide signals at a lower signal-to-noise (SNR) and sampling rates [16–18].

Diffuse correlation spectroscopy (DCS) is a related modality relying on the diffuse propagation of near-infrared light. DCS directly quantifies an index of blood flow (BF_i) by measuring the intensity fluctuations of speckles formed on the tissue surface due to injection of highly coherent laser light. These fluctuations are primarily driven by the motion of red blood cells [19]. Conventional DCS systems consist of a long coherence-length laser, single-photon counting detectors, such as single-photon avalanche diodes (SPADs), and photon detection time-tagging/correlation electronics to measure the temporal intensity autocorrelation function (g₂(τ)), which carries information about the tissue optical properties and microvascular blood flow. Standard single-wavelength DCS devices, however, are unable to disentangle the contributions of these parameters to the g₂ function and require either assumptions about tissue optical properties, or combination with FD or TD NIRS to obtain µ_a and µ_s’ and use them to quantify absolute values of the blood flow index (cm²/s) [17–20]. While hybrid FD-NIRS-DCS and TD-NIRS-DCS systems have been successfully used in adults and infants [17,18], the requirement for heavy fiber bundles on the NIRS side and bulkiness of the optical sensors prevent their use for extended periods of time in premature infants, making them impractical for long-term monitoring [21]. There is currently a need for a standalone, easy-to-use system capable of simultaneous quantification of cerebral BF_i (CBF_i), tissue optical properties and SO₂. Our group has previously demonstrated the possibility to adopt a multi-wavelength (MW) multi-distance (MD) DCS method for simultaneous quantification of flow, as well as absorption and reduced scattering coefficients in liquid phantoms using a benchtop experimental setup [22].

Following these promising results, we have now built a fully integrated, clinical ready MW-MD-DCS device optimized for simultaneous quantification of BF_i, SO₂, and CMRO_2i in preterm infants in the neonatal ICU. The instrument features three long coherence length laser modules emitting at different wavelengths in the near-infrared range, 8 single-photon detectors, and custom time-tagging electronics and aims at tackling current limitations of hybrid NIRS-DCS devices while still providing optimal performance, thus fostering a broader adoption of these techniques in the clinical field. While the original benchtop system and the new clinical ready systems have a similar hardware architecture, the new device includes new components and several features which improve light delivery and detection, a real time data acquisition and visualization, safety features, custom optical probes with integrated contact sensing and motion detection, packaged in a compact cart. With respect to the original device using human data, we have modified and optimized the fitting procedures for simultaneous quantification of BF_i and SO₂. In this work we report the clinical device and fitting method, the characterization and validations against a reference FD-NIRS instrument in phantoms and human subjects, and we show preliminary results obtained on a 32-week GA preterm and a 28-week ELGA infant.

2. Theory and methods

2.1 System design

A simplified block diagram of the MW-MD-DCS system is shown in Fig. 1. The system consists of three fiber-coupled long-coherence length (>10 m) lasers at 785, 808, 853 nm (iBeam Smart, TOPTICA Inc.) with a maximum optical power output of 80, 100, and 120 mW, respectively, with integrated laser drivers providing a sub-millisecond switching capability. While having a laser at a λ ≤ 765 nm would have provided better deoxy-hemoglobin concentration (HbR) sensitivity, commercially available lasers at these wavelengths cannot be switched on and off fast enough (∼1 ms needed) without reducing coherence length, have lower power (>50 mW needed), or like the distributed Bragg reflector (DBR) lasers used on our original device [22] have limited lifetime. The light from each laser is coupled into a separate 50/50 splitter (MMC-12-A-5050, Lightel), combined in two three 62.5 µm-core multimode fiber’s outputs and delivered to the probe in two illumination points. Light transmitted through tissue is collected at multiple SD separations (0.5, 1.5, 2.0, 2.5 cm) and delivered to two 4-channel silicon SPAD detector modules (SPCM-AQ4C, Excelitas Technologies Corp) using 9 µm-core few-mode fibers. While the larger fiber core diameter, relative to the typical 5 µm-core single-mode DCS fiber, results in a lower β value (∼0.15 vs 0.5 for single mode fibers), the increased number of photons results in better SNR [23]. The detector modules have a PDE of approximately 52% at 785 nm and 38% at 850 nm, with a dark count rate < 500 counts per second and an afterpulsing probability of 0.5% for a 50 ns hold-off time. Light at shorter separation (≤1.5 cm) is attenuated with variable optical attenuators (VOA) to prevent detector saturation and non-linear effects at high photon count rates (VOA780PM-FC, Thorlabs). The output of each of the detector channels is tagged using a 150 MHz clock using a custom FPGA-based (Field Programmable Gate Array) 20-channel time-tagging board. The FPGA additionally samples 8 auxiliary analog inputs at a 50 kHz sampling rate with a 14-bit resolution and ± 5 V input range. The auxiliary channels are used to collect and synchronize data from other clinical monitoring devices commonly used in a neonatal intensive care unit (NICU). The data collected are sent to a computer via a USB 3.0 interface board (EZ-USB FX3 USB3.0, Cypress Semiconductor Corp) for real-time display. storage and analysis. Finally, the system is powered at 12 V DC starting from the external 115 V AC 60 Hz input using a medical grade AC/DC converter (TPP 150-112, Traco Power).

 

Fig. 1. Simplified block diagram of the MW-MD-DCS system. Lines colored in red/blue represent the source/detection fibers, respectively.

Download Full Size | PDF

To precisely time-multiplex the three lasers, we made modifications to our existing electronics to provide FPGA driven trigger signals for the laser modules locked to the photon time tagging clock. The modified board outputs three separate transistor-transistor logic (TTL) trigger pulses, one for each laser. Each detected photon is additionally tagged with a laser ID to denote which source was active at the time. The lasers are operated at a 10 Hz repetition frequency with a 30% duty cycle, such that no two lasers are on at the same time. A 10% duty cycle ambient light acquisition phase is also included to perform background intensity subtraction for each individual laser cycle. The reduced duty cycle causes a √3 decrease in SNR at each wavelength, when compared to standard single wavelength DCS. The pulse duration, however, is sufficiently long to effectively fit the decay time of the measured autocorrelation functions (generally the decay occurs in the range from 10⁻⁵ to 10⁻⁴ s). We compensate for the reduced light power by using a double illumination strategy (2x increase in SNR) and an increased number of detectors at longer SD separations (√4 increase in SNR at 2.5 cm source-detector separation) to compensate for the decreased duty cycle and allow for in vivo measurements with a 3s sampling rate, as in [5].

The MW-MD-DCS system utilizes a custom lightweight contact-sensing probe [24], shown in Fig. 2(b). The probe (dimensions: 40 × 15 mm²) hosts the 3D-printed holders for source and detection fibers, a 3-axis accelerometer IC (ADXL327, Analog Devices Inc) on the top surface, and integrates a capacitive contact sensor (FDC2212, Texas Instruments) on the bottom surface. The capacitive contact sensor serves both as a safety feature to detect probe-detachment events and immediately shut down laser emission, and as a signal quality index, since proper probe attachment to the skin is pivotal for accurate optical measurements [24]. Light from the lasers is delivered to two illumination spots on the probe separated by 8.5 mm. In each spot the source fiber is connected to a holographic diffuser disk with a 40° diffusing angle and a 3.5 mm prism to meet ANSI standard skin exposure limits (ANSI Z136.1). Light is collected at 5, 15, 20, and 25 mm, four detection fibers and detectors are used at 25 mm, 2 at 20 mm, 1 at 15 mm and 5 mm, to improve SNR at larger separations. The 5 mm SD separation is not being used in the MW-MD-DCS fitting algorithm, but as a measurement of scalp blood flow, as done by Sunwoo et al [5]. To improve the ease-of-use and durability of the system, all fibers at the device side are terminated with 2 MTP connectors (Fibertronics), one for the 6 source fibers and one for the 8 detector fibers.

 

Fig. 2. (a) MW-MD-DCS system inside the modified medical cart. The system allows for storage of peripheral device used in a NICU, incorporates a laser interlock to the right of the device, and allows for placement of the laptop PC used to operate the device on the top of the cart, minimizing the overall system size. The fiber and auxiliary panel mounts are to the left below the laptop. (b) Top and bottom view of the contact-sensing double-illumination probe used in the MW-MD-DCS system. The locations of the light injection points are highlighted with red squares, whereas the detection points are highlighted in blue. The 3-axis accelerometer, placed on the top side of the probe, is highlighted in green, whereas the whole bottom surface of the probe makes up the capacitive sensor to constantly monitor the probe attachment to the patient. (c) Example of the double illumination probe placement on a preterm infant’s head.

Download Full Size | PDF

While the system is currently setup for unilateral measurements (3 sources in two locations and 8 detectors), the FPGA correlator and the software have the capability to handle up to 20 detection channels, so that by adding 2 more 4-channel detector modules and by further splitting the laser outputs, it can be upgraded for bilateral measurements.

2.1.1 Housing

The MW-MD-DCS system is housed in a customized portable medical cart (size 40 × 45 × 100 cm³), as shown in Fig. 2(a). The instrument is located in the bottom of the cart with a front facing panel containing the system power switches. The laser interlock switch is located on the right side of the cart for easy access in case of an emergency. The cart was designed with the goal of minimizing storage space and allowing for easy assembly and operation in the NICU. To prevent unwanted damage to the fibers during handling and improve ease of use, the electrical cables and fibers are routed through the back inside the cart and up to an opening on the top surface where the probe is connected via MTP connectors and routed to the subject using a retractable intravenous bag pole to prevent any tripping hazards. When not in use the optical probe, auxiliary electrical devices, and cables can be detached and placed inside the storage drawers.

2.1.2 Graphical user interface

We designed a new graphical user interface (GUI) that is primarily tasked with operating the hardware of the device, acquiring and visualizing data and results. Custom settings allow the user to configurate the number of lasers and detector channels, as well as adjust each laser power output. Additional settings allow to automate laser shut down based on thresholds for contact sensing, motion and detected light intensity. The raw data is acquired via the FX3 USB interface, and for the 3 lasers and 8 detectors a typical data rate is approximately 6 to 10 GB/hour and consists of photon timestamps, laser ID tags, auxiliary data, along with the system configuration. The GUI computes and displays data including g₂(τ) functions, photon count rates, raw BF_i for each channel and wavelength calculated using preset optical properties, and auxiliary channels. Additional features include the ability to add time-stamped notes, save data at predefined intervals, and show averaged signals.

2.2 Multi-wavelength multi-distance diffuse correlation spectroscopy

DCS measures the rapid temporal speckle fluctuations of light transmitted through tissue. The intensity autocorrelation function, g₂(τ), is related to the normalized field auto-correlation function g₁(τ) by the Siegert relation [25] in Eq. (1):

As demonstrated in our previous work [22] by assuming a homogenous medium, diffuse regime (µ_s’ >> µ_a) and uniform flow across larger SD separations (>10 mm) we can fit the MW-MD-DCS data (light intensity and g₂) across multiple λ, source-detector separation ρ, and τ to simultaneously obtain quantitative values of static (µ_a and µ_s’) and dynamic (BF_i) properties of tissue. The absorption coefficient is defined as the sum contribution of oxyhemoglobin (HbO), HbR, and water using Eq. (2):

To obtain the effective attenuation coefficient,

The MW-MD algorithm fits the MW-MD-DCS data (g₂ and intensity) over λ, ρ, τ to simultaneously solve for BF_i, a, b, HbO, HbR, all of which are independent of wavelength and SD separations by minimizing the cost function

To avoid errors due to local minima in the fitting procedure, we have implemented a multi-start approach. We first generate a set of initial guesses such that the starting point for each parameter is randomly selected from a uniform distribution within a physiologically relevant range for in-vivo measurements, and then fit the data using a Nelder-Mead simplex optimization algorithm [27]. Once the fitting is completed, we take the median value of each parameter across all runs.

For in-vivo data, to further reduce the noise in the autocorrelation and intensity values, we limit the fitting procedure to the initial baseline. The mean reduced-scattering coefficient, estimated during an initial stable period of about 10 minutes, is used to estimate blood flow and absorption for the rest of the measurement session, under the assumption that scattering remains constant. By using the initial scattering, µ_a is calculated by reversing Eq. (4) for effective attenuation and then calculating HbO and HbR using Eq. (2). The initial µ_s’, instantaneous µ_a, and measured g₂ functions are then used to fit for BF_i (as in Ref. [21]). This allows us to obtain more robust estimates of SO₂ and BF_i compared to performing the raw fitting method across the whole measurement, as well as a significant reduction (>30 fold) in processing times.

To detect stable periods in the measurement and reject motion artifacts we analyze the accelerometer and capacitive contact sensor data. We first calculate the moving standard deviation of the magnitude of the 3-axis accelerometer and determine whether a predetermined threshold value is exceeded. The threshold value for each of our probes is empirically determined by visually inspecting the accelerometer and intensity data. We additionally set a maximum threshold value for the contact sensor output, which if exceeded indicates poor contact with the subjects’ skin.

Finally, we calculate the cerebral metabolic rate of oxygen with the data acquired using Eq. (7) [28]:where f_v is the venous volume fraction (assumed to be 0.75 [29]) and SaO₂ is the arterial oxygenation.

2.3 Validation on liquid phantoms

To characterize the system ability to quantify µ_a, µ_s’, and the Brownian diffusion coefficient (D_B, the counterpart of BF_i when using a colloid suspension) we used tissue-like liquid phantoms. Specifically, we performed absorption and scattering titrations by progressively increasing concentrations of diluted black India ink and Intralipid, respectively, following dilution steps and optical properties used with the original benchtop device [22]. To change Brownian diffusion (as a stand-in for flow), since this device allows for faster sapling rates than the original benchtop system, we were able to use a temperature titration as in [18], which provides more linear step increases in D_B than when using a magnetic stirrer as done in [22]. For the three cases, the initial solution was made of 1000 ml of water, 40 ml of 20% Intralipid solution, and 10 ml of diluted ink. For the absorption and scattering titrations water was kept at ambient temperature (i.e., 21 °C).

For the absorption titration, we added progressive amounts (2 ml each step) of ink solution until the absorption coefficient increased by approximately 100% over the initial value. Conversely, for the scattering titration we added progressive amounts (4 ml each) of 20% Intralipid solution until µ_s’ increased by approximately 100% above the starting value. The liquid solution was manually stirred after every ink or Intralipid titration step, and the phantom was allowed to come to rest for about 2 minutes before taking a measurement to ensure the flow reached its baseline value. Measurements were performed simultaneously with the MW-MD-DCS system and a combined 8-wavelength FD-NIRS (MetaOx, ISS Inc., the DCS component of the MetaOx was not used for these phantom validation experiments) for validation [18]. Both instruments used immersion fiber probes with one source location and 4 detector locations at 15, 20, 25, and 30 mm separation from the source. The FD-NIRS wavelengths are 672, 690, 706, 726, 759, 785, 813, and 830 nm. For each instrument a total of three 30 s measurements were taken per each absorption or scattering titration step. To validate the D_B measure, we compared the D_B values obtained using the MW-MD-DCS algorithm with D_B calculated from the MW-MD-DCS autocorrelations, but using the optical properties recovered by the FD-NIRS system. Furthermore, due to the MW-MD-DCS and FD-NIRS having only the 785 nm wavelength in common, to compare the two devices the results of the FD-NIRS system were interpolated/extrapolated to obtain µ_a and µ_s’ values at 808 and 853 nm.

For the temperature titration, we first cooled the liquid phantom down to 5 °C. The phantom was then placed inside a larger container filled with icy water, which was then placed on top of a heated magnetic stirrer to ensure a uniform heat transfer between the external container and phantom. To minimize heat-driven natural convection effects, the heater level was adjusted to ensure the rate of increase in temperature was <1 °C/min. Temperature probes were used to monitor the temperature of both the phantom and the water bath. MW-MD-DCS data were acquired continuously for approximately 2 hours until the phantom temperature reached 28 °C. A FD-NIRS measurement was done at the end of the temperature titration to assess the optical properties of the solution. For validation, D_B of the solution with varying temperature was estimated as done in Ref. [18]. Due to the wide and largely unknown distribution of Intralipid particle sizes [30], however, the Stokes-Einstein equation for Brownian motion [31] was slightly modified to express D_B as a function of the measured flow at the start of the titration and temperature:

Since the phantoms did not have hemoglobin, Eq. (6) was slightly modified such that absorption was expressed as the sum of contributions from water and black India ink. The extinction coefficient of the diluted ink solution was obtained using a broadband illumination halogen lamp (HL-2000-FHSA, Ocean Insight) and a near-infrared spectrometer (STS-NIR, Ocean Insight). The contribution of the Intralipid solution to absorption at the 3 wavelengths used was found to be negligible and therefore excluded. The range of initial values for the fitting algorithm used in liquid phantom measurements covered the expected values across all experiments.

For all titrations, to calibrate the two devices, we independently measured the optical properties of the liquid phantoms at the end of the final titration by performing a single-detector translational FD-NIRS measurement by moving the detector fiber from 1.5 to 3 cm in 0.5 cm steps using a motorized translational stage. Data from both devices was acquired at a sampling rate of 10 Hz and fitted at a frequency of 1/3 Hz, such that 30 data points were averaged for each time point, with the exception of the temperature titration where the fitting frequency was reduced to 1/60 Hz.

2.4 Human experiments

To validate the system ability to recover absorption, scattering, and blood flow in vivo, we acquired MW-MD-DCS data on the forearm of 7 healthy adults (5 male, 2 female, ages 21-60, body mass index = 25.2 ± 4.2 kg/m²) while performing an upper arm cuff occlusion. This study was reviewed and approved by the Mass General Brigham Institutional Review Board (#2019P003074). All subjects gave informed consent prior to the start of the experiment. In each subject we acquired two sets of measurements with the MW-MD-DCS system and with the reference FD-NIRS-DCS system. For the MW-MD-DCS the probe had 5, 15, 20, and 25 mm SD separations, and the same long separations (15, 20, and 25 mm) were available for the FD-NIRS-DCS system. The probes were placed on the forearm muscle (extensor carpi ulnaris) in a marked location to minimize differences during 4 measurement’s repetitions. Each measurement consisted of a 60 second baseline followed by 4 steps of 30 mmHg cuff pressure increase, with each step maintained for 30 seconds. After the fourth step, at a pressure of 120 mmHg, the arm cuff was released, and data collected for an additional 60 seconds to allow for the return to baseline. The step change in pressure was used to test the system’s ability to measure gradual reductions in blood flow, as opposed to a complete arterial occlusion, which results in a null blood flow. Both devices acquired data at a rate of 10 Hz and for each step data were averaged over 3 seconds to estimate absorption, scattering, BF_i and SO₂.

2.5 Preterm infant measurements

This study was reviewed and approved by the Mass General Brigham Human Research Committee (Institutional review board #2014P002022) and was conducted at Brigham and Women’s Hospital. Preterm infants born at 26-32 weeks in a stable condition with no major congenital or genetic abnormalities were enrolled in the study. Parents were approached by the clinical staff who explained the study and obtained informed consent to start monitoring the newborn infant no later than 48 hours of life. Preterm infants were monitored in up to 6-hour periods for 3 consecutive days.

The optical probe was attached to the infant’s forehead while avoiding the midline section, as shown in Fig. 2(c), and held in place using a thin adhesive hydrogel layer in between the probe and the infant’s skin. The 3-axis accelerometer was used to detect motion and the contact sensing capacitance signal was used to detect possible probe detachments. In addition, the study operator monitored the infant via a remote camera system and logged events in the GUI. Heart rate and arterial blood oxygen saturation (S_aO₂) were co-acquired using a commercial pulse oximeter (Radical-7, Masimo).

Acquired raw photon timestamps were analyzed at a 1 Hz sampling rate, such that 10 complete laser cycles were averaged. The calculated g₂ and intensity data were then fitted using the MW-MD-DCS fitting algorithm at a 1/3 Hz rate. To increase SNR and reduce computing time, the fitting algorithm to estimate scattering was applied only to the starting baseline period measurement of 10 min duration. The resulting µ_s’ value was used to estimate µ_a, SO₂ and BF_i and CMRO_2i during the rest of the measurement session. Accelerometer data was used to detect periods of motion and remove motion artifacts during post processing. To automatically detect probe detachment during the monitoring period, the capacitive signal is initially calibrated to be ∼1.15 V. The laser shut down threshold is set at a value of 1.37 V and if the capacitance reaches that value the acquisition software triggers the laser shutdown to ensure subject’s safety.

3. Results

3.1 Validation on liquid phantoms

The results of the liquid phantoms titrations experiments are shown in Fig. 3, where from left to right we show the absorption, scattering, and temperature titration, and from top to bottom we show the absorption and reduced scattering coefficients at 785, 808 and 853 nm and fitted D_B measured with the MW-MD-DCS system and validated using FD-NIRS.

 

Fig. 3. Results on liquid phantom for absorption (a-c), scattering (d-f), and temperature titrations (g-i). Absorption and reduced scattering values obtained using the MW-MD-DCS system are shown using star symbols, while dashed lines or circle symbols represent results obtained with the reference FD-NIRS instrument. For the temperature titration, the dashed lines represent the values obtained by the FD-NIRS at the end of the measurement. D_B is reported in panels c, f and i; in figures c and f the FD-NIRS flow was estimated by taking the optical properties of the FD-NIRS and the g₂ information from the MW-MD-DCS and averaging the flow across all wavelengths, in panel i the (dashed line) Brownian diffusion coefficient for the temperature titration was calculated using Eq. (8). Plots a, b, d, e, g, h share the same legend (reported in d).

Download Full Size | PDF

During the absorption titration the estimated absorption coefficient changes linearly with the ink volume, whereas the reduced scattering coefficient and D_B remain relatively constant throughout the titration. The MW-MD-DCS absorption and scattering values are in good agreement with the values obtained with the FD-NIRS system with an average error < 7%. The average deviation between the MW-MD-DCS fitted flow and the flow estimated using FD-NIRS optical properties and the g₂ measured with the MW-MD-DCS system was 1.4 × 10⁻⁹ (cm²/s) corresponding to an average error of < 8%. For the scattering titration, the estimated µ_s’ increased linearly with the Intralipid volume, while µ_a and D_B remain relatively constant. The average error for µ_a and µ_s’ was < 12%. The average deviation for flow quantification using the FD-NIRS or the MW-MD-DCS optical properties was 2.1 × 10⁻⁹ (cm²/s) resulting in an average error < 12%. A second absorption titration on high absorbing phantoms was performed separately and results are reported in the Supplement 1 in Fig. S4. For the temperature titration we show the time traces for µ_a, µ_s’ and D_B estimated with the MW-MD-DCS during the 2-hour period, as well as the FD-NIRS µ_a and µ_s’ measured at the end of the titration experiment (dashed lines) and the estimated D_B based on the phantom temperature. The measured MW-MD-DCS D_B is in good agreement with the Brownian diffusion equation estimates from Eq. (8) with an R² = 0.96. The absorption and reduced scattering coefficient values remain relatively constant during the temperature titration.

3.2 Validation on adults

Figure 4 shows time traces for blood flow and oxygen saturation measured with the MW-MD-DCS and the FD-NIRS-DCS systems in 6 subjects. Subject 1 was excluded due to a problem with the pressure cuff. The µ_s’ value used for the iterative method was obtained by taking the average of the entire measurement. Despite the fact the MW-MD-DCS and the FD-NIRS-DCS measurements were not simultaneous, we observed a good agreement between the BF_i and SO₂ time traces obtained with the two modalities. Figures S5 and S6 show the linear regression and Bland Altman analysis between the two devices, where each dot represents the mean value measured each pressure modulation phase (excluding the return to baseline), where we discard the first 5 s and last 3 s. These results show a good correlation for both SO2 and BFi (slope = 0.71, R²= 0.66 for BF_i, slope = 1.02, R²= 0.61 for SO₂), with an average deviation between the two devices of 4.9% for SO₂ and 9.2 × 10⁻¹⁰ cm²/s (20%) for BFi.

 

Fig. 4. Average time traces for blood flow (left column) and oxygen saturation (right column) for each subject. Darker colored lines represent the mean, while the shaded area represents the standard deviation across 2 repetitions. Data acquired using the MW-MD-DCS system are shown in green, while data acquired using the FD-NIRS-DCS system are shown in red.

Download Full Size | PDF

3.3 Clinical data on older preterm infants

SO₂, HbT, and BF_i time-traces measured on a 32-week GA infant (female, birthweight = 1530 g) and a 28-week GA (female, birthweight = 1180 g) are shown in Fig. 5 and 6, respectively. Co-acquired pulse oximeter SaO₂ and HR are also shown. All data was analyzed at a frequency of 1/3 Hz using the iterative fitting method described in section 2.2. The first 10 minutes of the measurement were used as the baseline period to estimate a reduced scattering coefficient. For the 32-week GA infant the estimated reduced scattering coefficient was 5.8, 5.6, and 5.4 cm^-1 at 785, 808, and 853 nm, respectively, and 6.2, 5.9, and 5.6 cm^-1 for the 28-week GA infant. While for the infant population we could not directly validate the method against a gold standard, the estimated scattering, hemoglobin, flow and oxygen metabolism values are within a reasonable expected range [32]. Data from the contact sensor indicated proper attachment of the probe throughout the measurement even during periods of motion (highlighted in gray). Accelerometer and capacitive proximity sensor data for the 28-week and 32-week preterm infants are shown in Fig. S7 and S8, respectively.

 

Fig. 5. Example of a measurement performed on a 32-week GA preterm infant. Top panel shows the total hemoglobin concentration (purple line, right y axis) and SO₂ (red line, left y axis) values, second panel shows BF_i (blue line, left y-axis) and CMRO_2i (green line, right y-axis) at ρ = 2 cm, bottom panels show the SaO₂ and heart rate measured with a commercial pulse oximeter. Periods of the measurement where motion was detected are denoted using a shaded gray area. The calculated baseline scattering used in the iterative fitting was 5.8, 5.6, and 5.4 cm^-1 at 785, 808, and 853 nm, respectively. MW-MD-DCS data highlighted in gray and starting at t = ∼1:15 is missing due to a pause in the measurement.

Download Full Size | PDF

 

Fig. 6. Example of a measurement performed on a 28-week ELGA preterm infant. Top panel shows the total hemoglobin concentration (purple line, right y-axis) and SO₂ (red line, left y-axis) values, second panel shows BF_i (blue line, left y-axis) and CMRO_2i (green line, right y-axis) at ρ = 2 cm, bottom panels show the SaO₂ and heart rate measured with a commercial pulse oximeter. Periods of the measurement where motion was detected are denoted using a shaded gray area. The calculated baseline scattering used in the iterative fitting was 6.2, 5.9, and 5.6 cm^-1 at 785, 808, and 853 nm, respectively.

Download Full Size | PDF

4. Discussion

We have developed and characterized the first stand-alone MW-MD-DCS system capable of simultaneously quantifying cerebral optical properties, hemoglobin oxygen saturation, blood flow and CMRO_2i in infants, by utilizing the light intensity and autocorrelation function data across three wavelengths and three SD separations. We have compared the results to a commercially available hybrid FD-NIRS-DCS system. The values obtained on liquid phantoms show good agreement with the reference device and errors are comparable to those found in the literature [5,18,22]. In addition to the absolute values, titrations measurements followed the expected changes with a small fitting crosstalk between scattering and BFi. We reduce the influence of the uncertainty in the µ_s’ estimates in in-vivo measurements by assuming that scattering does not change and using a fixed µ_s’ coefficient, measured at the start of each session.

To demonstrate the system performance in retrieving accurate absolute BF_i and SO₂ values in-vivo we performed measurements on the forearm of healthy adults during an arm cuff occlusion experiment. The forearm was chosen due to our SD separations being optimal for preterm infants but too short for adult foreheads. Furthermore, measurements on the forearm allow for easy manipulation of blood flow using an arm cuff. We believe the decrease in accuracy (from 8% to 20% error in BF_i estimates), relative to liquid phantom measurements, can be attributed to several factors. First, measurements by the two devices were interleaved and performed sequentially which introduces variability across measurements. Slightly different positions and pressures of the probe could impact blood flow and blood volume. Furthermore, the arm cuff used to modulate the pressure was manually inflated, which can result in differences in cuff pressures by a few mmHg. Moreover, the heterogeneity of the forearm can be a significant source of error in the fitting procedure. It must be noted that the MW-MD-DCS system, like most CW NIRS devices, relies on the assumptions that the tissue being probed is homogenous, the superficial layer is thinner than 4 mm, as well as the light being collected is in the diffuse regime [33]. While multi-layer models have successfully been implemented in human adults [34], no models have been validated for the preterm infant population to the extent of our knowledge. This device was specifically designed for measurements on a preterm infants’ forehead and is not well suited for measurements on forearm muscles, which can be highly heterogenous due to the presence of a fat layer, muscle, and bones. The presence of heterogeneities can introduce distortions in the measured parameters and by extent the fitting algorithm. Despite these uncontrollable variances introduced by the experimental conditions, our device still demonstrated its ability to recover optical properties, BF_i, and SO₂ and the average errors of 20% and 4.9% for BF_i and SO₂, respectively, are considered acceptable.

Adding larger/more SD separations to the system would improve the µ_eff estimates, but the limited efficiency of the detectors, as well as the maximum optical power permitted by safety limits, would result in limited improvements at a much greater cost.

To demonstrate the clinical feasibility of our device, we measured stable preterm infants and show examples of measured oxygen saturation, total hemoglobin, BF_i and CMRO₂ time traces along with the heart rate and arterial oxygenation values obtained using a commercially available pulse oximeter. We show that during periods of little to no motion the device is stable and able to continuously measure SO₂ and BF_i, and the ability to assess the interference of motion artifacts. Furthermore, we tested the MW-MD-DCS custom sensing probe laser safety feature during preterm infant measurements and confirmed the system automatically shuts down the lasers well before the probe is fully detached from the infant’s forehead. Clinical data reported in this work were intended to only show clinical validation of the device: we are currently enrolling preterm infants at Brigham & Women’s Hospital with the goal of measuring a large number of subjects. Thorough statistical analysis of data obtained on ELGA infants will be reported once a sufficient sample size is available.

Like other NIRS devices, a limitation of the MW-MD-DCS system is its high susceptibility to motion as well as poor coupling between the probe and infant’s skin. Consequently, the use of the thin adhesive hydrogel layer, as well as the presence of lanugo hairs commonly found on preterm infants [35], can introduce large changes in the light coupling. This issue can further be exacerbated depending on the positioning of the probe, as well as the head geometry (curvature). To address some of these issues, we plan on redesigning the geometry of the probe in the future to reduce its overall size and allow for easier placement and attachment.

5. Conclusion

We have developed the first standalone MW-MD-DCS device for monitoring of premature infants, which is able to simultaneously recover optical properties (µ_a and µ_s’), SO_2, CBF_i, and CMRO_2i using quasi-CW illumination. The device was experimentally characterized and validated on liquid phantoms as well as in vivo on healthy human adults during arm cuff occlusion. We show examples of measurements on stable preterm infants, demonstrating feasibility of the method in a clinical setting. MW-MD-DCS has the potential to be used as a clinical tool for continuous long-term monitoring of preterm infants’ cerebral hemodynamics. The presented device is compact, portable, easy to use, and has multiple safety features to prevent harm to the infants, and by utilizing multiple wavelengths, the system provides more robust estimates of BF_i compared to conventional single-wavelength DCS systems. The device is currently used at Brigham Women’s Hospital to monitor ELGA infants with the goal of assessing initial clinical utility.