Laser Doppler blood-flow imaging (LDBFI) has been applied for many years to monitor microcirculation. It measures the frequency shift induced when light is scattered by moving red blood cells. The technique has been used in relation to many conditions, including burn and wound assessment and understanding of inflammatory responses and diabetes. Early systems monitored blood flow at a single point and then progressed to build up images through point-by-point scanning. However, acquisition time can be slow (up to five minutes for a full image), which means that images are susceptible to motion artifacts. In addition, dynamic blood-flow changes cannot be measured.
Laser Doppler (LD) signals have frequency components up to 20kHz, which is too high for detection by conventional video cameras. An alternative approach, laser speckle-contrast imaging1 measures the contrast of light scattered by tissue over the integration time of a conventional camera. A low speckle contrast indicates faster blood flow. Because the rapid fluctuations in light intensity cannot be measured by a conventional camera, this approach relies on an accurate model relating contrast to flow. Recent developments in high-frame-rate CMOS cameras mean that these intensity fluctuations can now be measured directly, which has enabled rapid acquisition of LD signals.2,3
There are still two drawbacks. First, a bottleneck exists in the data flow between the sensor and the external (off-chip) processor. This can be overcome by employing a smart-CMOS approach. Second, the large amounts of data passed from the sensor may result in additional congestion at the processor.2 We have addressed the latter by facilitating parallel processing on a field-programmable gate array (FPGA).
Smart-CMOS optical sensors are arrays of photodetectors that perform on-chip processing. In LDBFI, calculating the blood-flow value on-chip allows us to overcome the data bottleneck between sensor and processor. This approach removes the requirement for data transfer at a frequency of 40kHz for each pixel, since only the processed blood-flow value is transfered off-chip. In addition, processing can be tailored to typical signals because our sensor is custom-made. For example, LDBFI works on the principle of detecting small AC signals on a large DC pedestal (with a modulation depth of ~1%). Consequently, we amplify the AC component by a factor of 50 and apply unity gain to the DC voltage. Fewer bits are therefore required in the on-chip analog-to-digital converter.
The processing required to extract a blood flow measurement from a detected Doppler signal involves integration over a bandwidth of typically 200Hz to 20kHz, normalization by the DC level, passing the signal through a frequency-weighted filter, followed by squaring and time averaging.4 Processing is relatively simple. However, it is a challenge to implement this approach in silicon spanning multiple pixels to ensure that the fill factor remains high and the sensor inexpensive. The low frequency range of the data means that the size of on-chip capacitors and resistors can become very large. Designs are required that can implement the processing electronics spatially efficiently.
To date, we have designed and demonstrated a range of test sensors with different characteristics, all scalable to larger arrays.5,6 We show an example in Figure 1, consisting of a 16×1 photodiode array, current-to-voltage converter, AC amplifier, anti-aliasing filter, analog-to-digital converters, and digital filters. An example blood-flow signal from an occlusion and release test is illustrated in Figure 2. Our most recent design is a 64×64-pixel array, which builds on the 16×1 array design to provide a fully integrated 2D imager.
Figure 1. A 16×1 array of photodiodes uses on-chip processing to extract blood-flow information. SRAM: Static random-access memory.
Figure 2. Typical blood-flow signal from an occlusion and release test. a.u.: Arbitrary units.
We demonstrated our field-programmable gate array (FPGA) approach for faster off-chip processing in collaboration with Moor Instruments (UK). They developed a line-scanning system (moorLDLS) that can obtain 64 intensity values simultaneously using a linear photodetector array. A bottleneck at the processor limits the speed to one frame per ~12s. The FPGA provides parallel processing, enabling acquisition of a 64×64-pixel image in only ~4s (see Figure 3).
Figure 3. Blood-flow image of a healthy volunteer's hand using line scanning. Color scale: amplitude in arbitrary units. Processing is performed off-chip using a field-programmable gate array.
Smart-CMOS sensors overcome the data-transfer bottleneck, thus enabling full-field LDBFI. We next aim to take the 64×64-pixel system to the University of Southampton (UK) to image blood-flow changes in inflammatory responses. Continued improvements in camera technology will ultimately lead to development of cameras that can rapidly sample LD signals for external processing by, for example, FPGAs. Nevertheless, smart-CMOS sensors will still be useful in applications where a compact design is important, such as in capsule endoscopy. Higher-frequency applications of our FPGA-assisted approach include ultrasound-modulated optical tomography, imaging air flow, and vibrometry where frequencies can be in the megahertz range.
This research was funded by the UK Department of Health (New and Emerging Applications of Technology program) and the UK Technology Strategy Board. We are grateful to our collaborators at Moor Instruments, the Universities of Exeter and Southampton (UK), and the Nottingham University Hospitals Trust. We are also grateful to our academic colleagues (J. A. Crowe and Y. Zhu), postdoctoral researchers (D. He and X. Xu), and PhD students (N. Hoang and J. Himsworth).
Stephen Morgan, Barrie Hayes-Gill
Electrical Systems and Optics Research Division
Faculty of Engineering
University of Nottingham
Stephen Morgan is an associate professor and reader in biomedical optics. His research interests include laser Doppler blood-flow imaging, polarized-light measurements, ultrasound-modulated optical tomography, and sensing techniques for regenerative medicine.
Barrie Hayes-Gill is an associate professor with research interests in biomedical electronics and integrated-circuit design. He has published over 150 papers and holds six patents for medical electronic devices.