On-chip laser Doppler vibrometer for arterial pulse wave velocity measurement

Pulse wave velocity (PWV) is an important marker for cardiovascular risk. The Laser Doppler vibrometry has been suggested as a potential technique to measure the local carotid PWV by measuring the transit time of the pulse wave between two locations along the common carotid artery (CCA) from skin surface vibrations. However, the present LDV setups are still bulky and difficult to handle. We present in this paper a more compact LDV system integrated on a CMOS-compatible silicon-on-insulator substrate. In this system, a chip with two homodyne LDVs is utilized to simultaneously measure the pulse wave at two different locations along the CCA. Measurement results show that the dual-LDV chip can successfully conduct the PWV measurement.


Introduction
Pulse wave velocity (PWV) is an important measure for arterial stiffness, which is highly related with many cardiovascular diseases. A larger value of PWV indicates a greater arterial stiffness and thus a higher risk of cardiovascular events, so in clinical practice PWV is used as an important and independent predictor of cardiovascular mortality [1][2][3][4]. In practice, the carotid-femoral PWV, which describes the average velocity of the arterial pulse propagating from the carotid to the femoral artery, is usually used. The European Society of Hypertension suggested that a carotid-femoral PWV of 12 m/s can be considered as the threshold of an estimate of subclinical organ damage [5], and individuals with a carotid-femoral PWV higher than this threshold should seek for clinical treatments to prevent cardiovascular events.
Several devices have been introduced to measure the carotid-femoral PWV, the most well known being the Complior and Sphygmocor system [4]. For the assessment of local arterial PWV, however, there are no widespread commercial devices. It is only since recent that efforts have been undertaken to measure the propagation of the pulse using fast ultrasound-based techniques [6,7]. Additionally, the measurement setup of the carotid-femoral PWV is usually bulky and sometimes causes discomfort to subjects because they are palpated in the groin. A new noninvasive method to measure the local PWV of the common carotid arteries (CCAs) using laser Doppler vibrometry (LDV) technique has been proposed [3,8]. This method simplifies the measurement procedures and also overcomes the uneasiness in the conventionally used carotid-femoral PWV measurement. However, the proposed measurement setup for the CCA PWV measurement is still bulky, which prevent this technique from being widely used.
In this paper, we present a compact device for the CCA PWV measurements, using a photonics integrated circuit (PIC) consisting of two individual homodyne LDVs [9]. So far, not many LDV PICs have been reported. Compared to the compact self-mixing technique [10], onchip homodyne LDV technique has easy demodulation algorithms. The footprint of this chip is smaller than 5 mm 2 . This dual-LDV PIC is realized on the silicon photonics platform using CMOS compatible technologies [11]. Silicon photonics has become very important in recent years to implement high bandwidth transceivers for datacommunication applications. With this technology one can integrate passive optical waveguides, optical filters, optical modulators, optical detectors and optical sources on a silicon-on-insulator (SOI) wafer and do so with the standard technology portfolio of an advanced CMOS fab. But recently the same technology has also gained considerable interest for sensing and biosensing applications [12]. The big strength of silicon photonics over other optical integration technologies is that one can make use of the very mature toolset in an existing infrastructure and that there is a direct route to volume manufacturing at low cost.
In the following part, the working principle of the proposed dual-LDV system is demonstrated first. Then we discuss the algorithm for the PWV measurement, especially the crosscorrelation method. The measurement results, including the calibration measurements of the dual-LDV PIC and a number of in-vivo PWV measurements, are demonstrated afterwards.

Description of the dual-LDV PIC
A detailed explanation to the principle of an on-chip single-point homodyne LDV has been given in [9]. The PIC-based PWV measurement device proposed in this paper consists of two such homodyne LDVs. The schematic of the on-chip system and the measurement setup is shown in Fig. 1(a). Despite the capability of integrating lasers and photo-detectors (PDs) on SOI chips [13,14], we still use external laser and PDs in this proof-of-concept setup to solely test the performance of the passive optical system. The light source and the PDs are connected with the chip via a fiber array (FA), in which 12 single-mode fibers (2 meter long, core/cladding = 9/125 μm) have one of their ends assembled on a V-groove array with a pitch of 127 μm.
During the operation, coherent light generated in the external laser is sent to the chip via the FA. An optimized optical coupling between the fiber and the on-chip grating coupler (GC) is ensured by a polarization controller (PC). In the 1×4 optical splitter (os1) light is split into four parts: two parts are sent to the measurement arms of the two stand-alone LDVs, the third part is split again into two reference signals with another optical splitter (os2), and the fourth part is directly sent out to a detector (Mon in Fig. 1(a)) to track the variations of the light intensity. Two light transmitting antennas (TAs) and two light receiving antennas (RAs), realized by means of grating couplers, are set in the measurement arms to send light out of the chip and receive back reflections, respectively. A lens is set in front of the chip to focus the two light beams from the TAs onto two locations on the neck region of the CCA. The two TAs (or RAs) have a separation S o = 0.98 mm to ensure an adequate large distance S i between the two measurement positions, which can be calculated as follows where M is the magnification of the optical system, d o and d i are the object and image distances, respectively. The back-scattered signals are collected by the same lens and focused back to the chip, where they are picked up by the RAs. To ensure that enough reflection is coupled to their corresponding RAs, the locations of the RAs are set very close the TAs. According to the Doppler effect, the recaptured reflection signals carry the information of the vibrations in their time-dependent frequency shifts. They are mixed with the reference signals in the two 90 • optical hybrids [15] (see Fig. 1(a)). From each 90 • optical hybrid, four optical signals are obtained and sent to two external balanced PDs. The footprint of this PIC is smaller than 5 mm 2 , and most of the space is taken by the FA. A photo of the chip is shown in Fig. 1(b), in which the actual area of the dual-LDV PIC is shown in the green box.
With the help of an analog-to-digital converter (ADC), the generated photocurrents of the balanced PDs are transformed into digital signals, which are used to recover the instantaneous velocity of each vibration using a digital "arc-tangent" demodulation method [9]. After the demodulation, two curves representing the instantaneous velocities of the vibrations can be obtained. The sampling time of the ADC t s determines the maximal detectable displacement speed of the neck surface v m = λ 0 /2t s , where λ 0 is the wavelength of the light. In our measurement, the sampling frequency f s = 1/t s is chosen to be 40 kHz, which corresponds to a v m of 31 mm/s. Note that the value v m is not related to PWV.
Considering the resolution of a measured PWV, it is better to use a large S i value, which is, however, limited by the size of the neck. In practice, the distance S i is normally chosen to be larger than 1 cm, indicating that the value of |M| is normally larger than 10. In this case, however, the numerical aperture of the lens on the image side is very small, which results in a strong reduction in the power of the backscattered light. In order to obtain a stronger reflection, two pieces of retro-reflective tapes are attached on the sites where the vibrations are measured. Considering the source side of the lens, the aperture of the lens D should be large enough to collect both light beams from the chip to avoid additional loss, indicating that D > D c = S o + 2d o tan(θ /2), where θ 11 • is the divergence angle of the output light (see Fig. 1(a)) and D c is the total size of the two beams when they reach lens. In this setup, a lens (C280TME-C from Thorlabs) with a focal length f = 18.4 mm and aperture D = 5.5 mm is used, and its position is adjusted with the help of a translation stage. The separation S i is around 1.5 cm, which leads to d o = 300 mm and D c = 4.8 mm (being smaller than D).
Since optical frequency shifters are not used in homodyne LDVs (see [16]), homodyne LDVs have a relatively low production cost compared to heterodyne. Hence homodyne LDVs are used in this device. However, homodyne LDVs also have drawbacks, e.g. they may not work well for vibrations smaller than half a wavelength. But this problem is not critical in the PWV measurement and can be alleviated by stabilizing the laser source and using a temperature controller. In addition, the wave pulse is usually very strong and thus weaker vibrations can usually be ignored.

Algorithms for retrieving PWV
The PWV is usually retrieved by measuring the transit time of the pulse wave between the two measurement locations, denoted as Δt. The value of PWV can be obtained from Δt and S i using the following formula The accuracy of the PWV measurement thus depends on the accuracy of the retrieved transit time Δt. The maximal detectable PWV value PWV m is ultimately limited by the sampling frequency t s of ADC, and their relation can be written as PWV m = S i /t s . Providing f s = 40 kHz and S i = 1.5 cm, a PWV m of 600 m/s can be obtained. However, due to the inaccuracy in the algorithm for retrieving Δt and deviations in the measurements, the PWV m is determined by the transit-time error t d in the measurement rather than t s . The value t d depends on many factors, including shape changes between the two pulses, errors in the LDV outputs due to the weak useful reflection and the existence of on-chip spurious reflections, and limitations in the algorithms used for retrieving Δt. In [3], a cross-correlation algorithm is used to retrieve the PWV, in which the transit time between the two measurement locations (Δt) is obtained from the cross-correlation of the two measured velocity curves. The accuracy of this method depends on the shape changes between the measured velocity curves of the two vibrations. These shape changes are partly caused by propagation dispersion of the pulse wave and reflection at sites with impedance mismatch [6], and partly caused by errors of homodyne LDVs when light reflection is not strong enough. Hermeling et al. proposed to use the transit time of the systolic foot of the pressure waveform [6], in which the deviation also depends on the shape variation and the accuracy of the LDV system. However, this algorithm is more complex than the cross-correlation method in many cases.

Measurement results
The LDV PICs are fabricated via ePIXfab [17]. A calibration measurement was firstly done on two separate earphones, on which two pulses were generated by two voltage signals with the same shape but a relative preset time delay Δt. The input light with a wavelength of 1550 nm has a power of 6 dBm and linewidth of 150 kHz. Light at this wavelength has the minimal loss in both optical fibers and SOI devices. These challenging test conditions correspond to a pulse wave velocity of up to 150 m/s, which is well beyond the values that can be expected in vivo. In Fig. 2(a), the velocity curves of the two pulses with a preset delay Δt = 0.8 ms is shown. It can be seen that a time delay of 0.8 ms is retrieved from the leading edge delay of two pulses, while the delay derived from the cross-correlation method is around 0.85 ms (see Fig. 2(b)). The retrieved results of these two methods have been compared using a series of measurements (see Fig. 2(c)). The measured root-mean-square deviations of the leading-edge and cross-correlation algorithms are 0.13 ms and 0.08 ms, respectively. The cross-correlation method is less sensitive to noise and spurious reflections in this situation. Besides, their influences are also strongly reduced by filtering. Therefore, the estimated deviation is mainly caused by the shape difference between the two pulses. Then the device was used to measure the CCA PWV in vivo. The person under test was a healthy male volunteer aged 26 years. During the test, the volunteer was sitting in an upright position in front of the device, with two retro-reflective tapes attached to two locations along the CCA (see Fig. 3). The separation of the two locations was between 1.5 cm and 1.7 cm. The two light beams were adjusted to be focused on the two retro-reflective tapes, respectively. The demodulated velocity results for a representative measurement are shown in Fig. 4(a), and the cross-correlation of the velocities is obtained and shown in Fig. 4(b). It is seen that the transit time of the pulse is around 3.3±0.1 ms, and this corresponds to a PWV between 4.85±0.45 m/s. The displacement and acceleration signals are also shown in Fig. 5(a) and Fig. 5(b). If a proper filter is applied, the cross-correlation method can also work well with the acceleration signal, but not with the displacement signal, which has a strong low-frequency variation. Another byproduct of the LDV system is the heart rate of the subject. In this measurement, the average   heart rate of the subject is around 78 bpm, which is calculated according to the time delay between two adjacent pulses.

Conclusion
In this paper, the proof-of-concept of a miniaturized PWV measurement device utilizing dual-LDV photonic integrated circuit is demonstrated with initial experimental results. The crosscorrelation method is used to derive the PWV, and the time delay error was found to be 0.1 ms. When the separation between the two measurement spots is only 1.5 cm apart from each other, the measurable range of this device can cover all possible values of the CCA PWV. In future work, efforts should be made to integrate lasers and photo-detectors on the same photonic integrated circuits. A PIC with more than two LDVs is also interesting for future study since it can be used to obtain more information of the pulses at different locations.