Full-field laser Doppler blood-flow imaging and monitoring with CMOS image detector

We are developing a system for parallel, full-field, laser Doppler blood flow (perfusion) imaging and monitoring. This new imaging system allows obtaining 2D flow maps or to monitor flux signals from a plurality of separate predetermined points simultaneously with a 2D matrix of photodetectors; CMOS image sensor in our case. Until now we have achieved imaging time of 5 seconds for 64x64 pixels image, which is 4 times quicker than for conventional scanning laser Doppler imager. We believe that using this new system would decrease patient discomfort and produce clinical diagnosis. A description of the technique and current results are presented below.

Introduction

Laser Doppler Perfusion Imaging (LDPI) is an interferometric technique successfully used for visualization of two-dimensional (2D) microvascular flow-maps in a number of clinical settings including investigations of e.g. peripheral vascular disease, skin irritants, diabetics, burns and organ transplants. The technical principle is based on the Doppler effect wherein the light scattered by moving particle, e.g. blood cell, obeys a slight frequency shift, which can be measured by the heterodyne detection [1]. A 2D flow map is obtained by means of sequential measurements from a plurality of predetermined points. In classical LDPI systems this is achieved by scanning the area of interest with a narrow collimated or focused laser beam [2,3]. However this approach is time-consuming and suffers from artifacts coming from mechanical steering of the laser beam. Fujii et al. [4] and Briers et al. [5] have proposed an alternative full-field flow imaging technique using speckle contrast analysis, which is essentially a time-domain technique. Nevertheless working in the frequency domain allows to access more information [6]. Recently Serov et al. [7] suggested a new approach on parallel laser Doppler imaging: a true-random-addressing CMOS image sensor was used to detect Doppler signal from plurality of points on the sample illuminated with a divergent laser beam. Here the mechanical scanning is substituted by the photoelectrical scan resulting to a faster acquisition speed.

We further improved this concept of parallel LDPI based on a new generation of true-random-addressing CMOS image sensor. Our approach takes the advantages of the CMOS image sensor pixel architecture coupled with a digital signal processor (DSP) to improve the flow map refresh rate. We optimized the system performance in terms of sensitivity, accuracy, acquisition speed, and computation time.

Experimental set-up

A schematic of the set-up for parallel laser Doppler flow imaging is shown below. For the area illumination we used a 3mm diameter multimode optical fiber made of acrylic polymer. Light from a Diode Pumped Solid State (DPSS) laser (671nm, 50mW output power) was coupled to the fiber. The distal fiber end was imaged on the sample via a magnification lens. As the fiber is highly multimode the speckle pattern caused by the time-spatial coherence of the laser light was not resolved by the camera. Also, the intensity distribution on the fiber facet was uniform. Thus a homogeneous illumination of the area of interest was achieved. The size of the illuminated area, several centimeters in diameter, could be varied with the magnifications lens. For observation of the sample and for detection of the Doppler signal an intelligent CMOS camera was used, iMVS-155 (Fastcom Technology SA, Switzerland). This camera is based on the CMOS image sensor Fuga1000 from FillFactory (Belgium). It comprises of a matrix of 1024x1024 pixels with a pixel size of 8x8mm² and a fill factor of 70%. The analog-to-Digital Converter (ADC) has a sampling rate of 10MHz transforming the analog signal to digits at 8-bit resolution.

The electronic architecture of the sensor provides us with a possibility to read out a Window Of Interest (WOI) at higher frame rate - several kHz, depending on WOI size. Fast pixel acquisition allows the detection of the intensity variations induced by the Doppler-shifted light. The signal variations at each sampled pixel are stored in the camera memory (4.5MByte in total) and thereafter processed with a 32-bit floating-point processor (ADSP 21061 SHARC, 40MHz, Analog Device). The signal processing comprises the calculation of the zero- (M₀) and the first-moment (M₁) of the spectral power density S(v) of the intensity fluctuations I(t) on the detector. The average concentration, <C>, of moving particles in the sampling volume can be derived from the zero-moment of the power spectrum; and the first moment, which is in laser Doppler flowmetry called flux or perfusion, is proportional to rms speed of moving particles, Vrms, times their average concentration [8]. We calculated the power spectrum using FFT algorithm applied to recorded intensity variations of the signal at each sampled pixel of WOI. Thereafter both the perfusion and the concentration maps over an area or instantaneous variations of the perfusion signal in a chosen point of interest are calculated and shown on a display.

Experiments and results

Flux response test: velocity and concentration

For the flux response performance test a special artificial sample has been built of white Teflon whose optical properties matches well the optical properties of human skin. This plastic block contains a circle channel of 1 mm in diameter, placed at a depth of 1mm below the illuminated surface. The flow through the channel was varied in a controlled manner with a liquid pump in the velocity range from 0 to 3mm/sec. As a liquid we used an Itralipid™ solution at different concentrations: 0.1%, 0.5%, and 1%. The sample was illuminated with a 50mW laser beam over an area of 20mm in diameter. Doppler signal was recorded with 40kHz sampling frequency from a single pixel and processed thereafter.

In Fig.2 the dependence of the CMOS imager flux response to the velocity is shown for three different Intralipidä concentrations. The diagram shows a linear response of perfusion due to flow velocity up to 3mm/sec. The measured flux response to concentration demonstrates also an increase in the flux value. Flux response of the imager is not at zero for the zero velocity of flow due to the Brownian motion of the scattering particles in Intralipid™.

In vivo performance: Monitoring of blood flow

To monitor the perfusion in time the Doppler signal was measured from several pixels read-out at 40kHz sampling rate. This perfusion was measured on a fingernail illuminated by a 50mW laser beam of 40mm in diameter. The measured perfusion time traces obtained from a single pixel are shown. The left time trace demonstrates the perfusion response to the occlusion of the upper arm. Since the inflation of the occlusion cuff takes a certain time the perfusion decreases slowly after the occlusion until biological zero is attained. Once the occlusion is released the finger is quickly reperfused and the perfusion signal recovers the initial value. Also we measured the response of the perfusion when the subject performed a deep breath test. The measured perfusion time trace is shown on the right. Directly after the deep breath-in the perfusion decreases and after about 10 seconds it returns to the initial value. The perfusion is reduced since a deep breath induces a narrowing of the peripheral blood vessels; so-called vasoconstriction.

In vivo performance: 2D perfusion imaging; occlusion

We obtained the flow maps of the perfusion in the finger before, during, and after occlusion of the arm artery. An inflatable cuff, used for blood pressure measurements, is pumped to stop the blood flow. The finger was illuminated with a 40mm diameter laser beam of 50mW optical power. An area corresponding to 256x256 pixels was scanned in photoelectrical meaner with a sub-window of 64x8 pixels size. Thus the sampling rate for each pixel was 16.8kHz; 256 samples were obtained for each pixel. In total, acquisition, signal processing and data display took approximately 90 seconds to obtain one perfusion image of 256x256 pixels size. As expected, there is a decrease of the perfusion during the occlusion. After the occlusion is released the local perfusion raised above the initial value; this effect is known as reactive hyperemia.

In vivo performance: 2D perfusion imaging; cooling down cycle

Here we show the response of the finger perfusion submitted to a cooling down cycle. During two minutes the finger was held in ice water. Then we measured the perfusion maps directly after, 3 minutes after, and 10 minutes after the immersion into the cold water. The low temperature has the effect of decreasing the perfusion. After 3 minutes an increase in perfusion compared to the initial state is observed. After 10 minutes the perfusion attains its initial state.

Discussion

The results demonstrate the reliable performance of the new parallel laser Doppler imager applied to human skin. New laser Doppler system provides 2D perfusion maps as well as monitoring of the perfusion at several points over the illuminated area simultaneously. The system shows a linear response to the velocity up to 3mm/sec with an illumination power of 50mW over an area of 20mm in diameter. Actually the upper frequency cut-off is limited not only by the sampling rate of the sensor but also by the pixel response time before it can be read-out. This behavior has its origin in the non-integrating nature of the sensor: less current is needed for bright illumination. The stronger the light the shorter the pixel response time is [9]. Thus the bandwidth also depends of the intensity of the detected backscattered light. In our case the intensity was 0.16mW/mm² that is even 10 times lower than typically used in the traditional scanning imagers. The lower frequency cut-off is limited by the amount of memory available for storage the acquired sub-frames. E.g. in the measurements presented in this paper we had a bandwidth from 40Hz to 20kHz for monitoring the flow, and from 66Hz to 8.4kHz for the perfusion imaging.

The refresh rate of the perfusion images was about 90 seconds for 256x256 pixels. This time includes the acquisition, signal processing, and sending the data to the display. Most of the time is taken by FFT calculations. For comparison, the declared scan speed of a commercial laser Doppler imaging system MoorLDI™ (Moor Instruments Ltd, UK) is typically 20 seconds for 64x64 pixel resolutions, and 5 minutes for 256x256 pixel resolution. This is 3-4 times slower than for our imaging system. However the scanning imager can measure the areas of up to several detsemetres (50x50cm²) in size, while with the area illumination approach the imaged area is limited to several centimeters due to maximal power of the source laser. Another challenge on the way to improve the performance of our imager is the logarithmic intensity response, fixed-patter-noise, and 8-bit resolution, all of which reduce the SNR of the system. The effect of the logarithmic response is a very high dynamic intensity range (120dB) but unfortunately also a presence of high frequency noise, and Fixed Patter Noise (FPN). If the first challenge is essentially a property of the true-random-addressing pixel architecture, the last two can be obviously improved in new versions of the sensor.

Outlook

The CMOS image sensor market is developing very fast. New sensors with improved photoelectrical parameters are already available. Obviously this is of interest to test their performance for the parallel laser Doppler imaging. We believe that with newer CMOS image sensors better quality 2D flow maps would be obtained. Also, the refresh rate of the flow images can be increased using a faster DSP integrated to the camera, or even accomplishing the signal processing in a host PC having a fast communication with the camera.

Definitely the main advantage of the parallel imaging approach is a faster operation speed, which could allow observing high-resolution perfusion images with a refresh rate of less than a second.

We are aiming to extend our technique to dermatology, ophthalmology, and neurosurgery.

References

1. A.P.Shepherd, and P.A.Öberg, Laser-Doppler Blood Flowmetry, Kluwer Academic Publishers, Boston, 1990.
2. K.Wårdell, A.Jakobsson and G.E.Nilsson, "Laser Doppler perfusion imaging by dynamic light scattering", IEEE Trans. on Biomed. Eng. 40, 309-316 (1993).
3. T.J.H.Essex and P.O.Byrne, "A laser Doppler scanner for imaging blood flow in skin", J. Biomed. Eng. 13, 189-193 (1991).
4. H.Fujii, K.Nohira, Y.Yamamoto, H.Ikawa, and T.Ohura, "Evaluation of blood flow by laser speckle image sensing. Part 1", Appl. Opt. 26, 5321-5325 (1987).
5. J.D.Briers, G.Richards, and X.W.He, "Capillary blood flow monitoring using laser speckle contrast analysis (LASCA)", J. Biomed. Opt. 4, 164-175 (1999).
6. J.D.Briers, "Laser Doppler, speckle and related techniques for blood perfusion mapping and imaging", Physiol. Meas. 22, R35-R66 (2001).
7. A.Serov, W.Steenbergen, F.F.M. de Mul, "Laser Doppler perfusion imaging with a complimentary metal oxide semiconductor image sensor", Opt. Lett. 25, 300-302, (2002).
8. R.Bonner and R.Nossal, "Model for laser Doppler measurements of blood flow in tissue", Appl. Opt. 20, 2097-2107 (1981).
9. E.R.Fossum, "CMOS image sensors: electronic camera-on-a-chip", IEEE Trans. on electron devices 44, 1698-1698 (1997).