Portable System for Time-Domain Diffuse Correlation Spectroscopy

Abstract

We introduce a portable system for clinical studies based on time-domain diffuse correlation spectroscopy (DCS). After evaluating different lasers and detectors, the final system is based on a pulsed laser with about 550 ps pulsewidth, a coherence length of 38 mm, and two types of single-photon avalanche diodes (SPAD). The higher efficiency of the red-enhanced SPAD maximizes detection of the collected light, increasing the signal-to-noise ratio, while the better timing response of the CMOS SPAD optimizes the selection of late photons and increases spatial resolution. We discuss component selection and performance, and we present a full characterization of the system, measurement stability, a phantom-based validation study, and preliminary in vivo results collected from the forearms and the foreheads of four healthy subjects. With this system, we are able to resolve blood flow changes 1 cm below the skin surface with improved depth sensitivity and spatial resolution with respect to continuous wave DCS.

Fig. 1.

Example of pathlength distributions (dashed areas) of the photons at a sampling time of 1.5 ns. The optical properties used for the sample response function (SRF) are μ_a = 0.05 and μ_s’ = 10 cm⁻¹. (a) Two Gaussian IRFs with 100 ps and 500 ps FWHM, centered on t_s, and the corresponding pathlength distributions. (b) 500 ps Gaussian IRF with the addition of diffusion tails representative of two different detectors and the corresponding pathlength distributions. The IRFs are flipped so that the total delay (sum of the ToF and emission time relative to IRF peak) equals t_s to for the photons in the distribution. Note that the longer tail causes increased contribution of photons with short ToF to the pathlength distribution at t_s.

Fig. 2.

Block diagram of the system setup, where three MPD-RE detectors are used to measure the IRF and the light collected at 1 and 2 cm away from the source. A MPD-FG module collects the light at a 1 cm separation from the source. For the laser source, we chose the Vis-IR 500 because of the longer coherence length. A correlator board acquires the ToF measured by TDC-cards and pairs this information with the absolute arrival time of the photon, then transfers the data to a PC via USB 3.0. (b) A photo of the system, showing an enlarged view of the optical probe (the probe has also a 15 mm separation, not used in this work).

Fig. 3.

Autocorrelation functions (g₂) of the CW equivalent photons (i.e. using the whole TPSF) when using a 12.5 MHz laser repetition rate (fL), 150 MHz correlator sampling frequency (fS), and 1 cm source-detector separation. For these measurements, we used a silicone oil liquid phantom with a very slow decay (in the 100 μs to 1 ms range) to achieve a constant pattern in the μs range and better visualize the oscillations. The g₂ curve computed without any correction shows oscillations which are due to the fact f_L is not equal to nor an integer multiple of f_S. By applying either a jitter or a resampling correction the oscillations are completely removed.

Fig. 4.

Pulse width comparison (a) and g₂ comparison (b) of the four lasers tested. The g₂s are evaluated on a silicone oil liquid phantom at 1 cm source-detector separation, while selecting the photons in a gate of 60 ps centered at the TPSF peak. For narrower laser pulses, the g₂ are much lower because of the shorter coherence length.

Fig. 5.

Beta measured at different gates centered on the TPSF peak. We changed the gate width from 10 to 500 ps (roughly the FWHM of the VisIR-500 laser). The beta drop changes slope around 180 ps, that corresponds to the coherence length of the laser (~40 mm).

Fig. 6.

IRF comparison using the two detectors. The MPD-FG IRF (green) has a fast extinction time of 585 ps to drop two orders of magnitude, while the MPD-RE IRF (blue) takes about 2.28 ns. A measurement of the VISIR-70 and VISIR-500 lasers (blue vs red curve) both acquired with the MPD-RE shows the effect of the source width on the long diffusion tail. For an increased pulse width, the diffusion tail contribution starts higher (at −13 dB instead of −18 dB).

Fig. 7.

Normalized count rate (black, left y-axis), IRF peak position (blue, right y-axis), and IRF FWHM (green, right y-axis) for a long acquisition time. The system stabilizes after about 25 minutes of warm-up time.

Fig. 8.

Results of the two-layer experiment. On the TD-DCS side, the g2 of the early (EP, 240 to 120 ps before the TPSF peak) and late (LP, 360 to 720 ps after the TPSF peak) photons show a large decay time difference between the Intralipid (IL) and the silicone oil (SO) phantoms, while the normalized TPSFs show minor differences due to the small difference in optical properties. For all detectors, the g₂ of the 20 mm top-layer is close to the silicone oil one. When selecting the late photons, the g₂ of the 0.5 and 1 cm top-layer phantoms show a decay closer to the Intralipid. On the CW-DCS side, the difference in the g₂ decay between intralipid and silicone oil is evident at all source-detector separation. For the two-layer phantoms, the g₂ show a decay closer to the one of the bottom layer whenthe top layer is thin and the source-detector separation is larger.

Fig. 9.

Comparison of the normalized BF_i at different thickness between TD- and CW-DCS. TD-DCS with LP at 2 cm separation shows the maximum sensitivity to depth, while LP results at 1 cm are comparable to CW-DCS at 2 cm. The sensitivity is good up to 15 mm layer thickness, and becomes poor at 20 mm.

Fig. 10.

Comparison of the normalized BF_i (average and standard deviation obtained by splitting the measurement in 60 parts of 10 s each) at different thickness between the MPD-FG (a) and the MPD-RE (b) detectors at 1 cm and the MPD-RE at 2 cm (c) separation when pushing the selection of photons to late gates, up to 1 ns after the TPSF peak. The sensitivity of the MPD-FG increases while the MPD-RE at 1 cm starts losing sensitivity at thin layer due to increasing contribution of the diffusion tail. The MPD-RE at 2 cm is less penalized by the detection tail thanks to the slower decay of the TPSF at larger source-detector separation, but the sensitivity at very-late and ultra-late gates (VLP and ULP) is not improved as in the MPD-FG detector.

Fig. 11.

Results of the flow experiment when the source-detector separation is 1 cm (a) and 2 and 3 cm (b). The highest contrast is achieved when the tube is centered between source and detector: x = 5 mm (a), x = 10 mm (b) and x = 15 mm (b, orange curve). At 1 cm source-detector separation, the flow is not detected by the early photons (EP), while the maximum flow contrast is obtainable when selecting the late photons (LP) and using the short tail detector (MPD-FG, brown line). Similar results are obtained at 2 cm source-detection separation. CW-DCS at 3 cm (orange) provides lower contrast and spatial resolution than LP TD-DCS.

Fig. 12.

Results of the cuff occlusion experiment. Cuff pressure shown in panel (a), blood flow index when selecting early photons (b) and late photons (c) at 1 cm source-detector separation. In both cases, and with both detectors, we see BF_i drop to almost zero during ischemia. At late gates, the higher efficiency of the MPD-RE (blue curve) results in less noise than the MPD-FG (brown curve). The bold lines are means and faint lines the standard deviations of eight trials in total.

Fig. 13.

Relative BF_i changes at 1 cm source-detector separation using the MPD-RE (blue) and the MPD-FG (brown) before and during the application of light manual pressure on the optical probe placed on the forehead. By using the early photons, the BF_i drops by 31% after applying the pressure, as expected. By using the late photons, the MPD-FG shows a small drop of 7%, while the MPD-RE shows a drop of 22% due to the tail contribution.