Microvascular imaging: techniques and opportunities for clinical physiological measurements

Laser Doppler flowmetry (LDF) uses the frequency shift of low-power (typically in the region of 1 to 5 mW and single mode) laser light produced by the Doppler effect to assess tissue blood flow (Oberg et al 1979). The initial development was single point measurement technology using fibre-optics, with developments in imaging coming to fruition in the early 1990s (Essex and Byrne 1991, Wårdell et al 1993). Depending on the laser wavelength employed, LDF can assess blood flow in the nutritional capillaries close to the skin surface and the underlying arterioles and venules deeper in the dermis involved in regulation of skin temperature (figure 1). LDF could also be used for other tissues when exposed at surgery, e.g. colonic blood flow (Hajivassiliou et al 1998). The tissue thickness sampled by a red or infrared laser system is typically 1 mm, the capillary diameters 10 µm and the erythrocyte velocity range 0.01 to 10 mm s⁻¹. Blood flow information can also be available for the larger elements of the microvasculature, i.e. 100 µm, although LDF underestimates flow when these vessels are vasodilated due to system bandwidth limitations.

Zoom In Zoom Out Reset image size

The technique depends on the Doppler principle whereby low power light from a monochromatic stable laser e.g. a 633 nm helium neon gas laser or a single mode 670 or 780 nm laser diode, incident on tissue is scattered by moving red blood cells and as a consequence is frequency-broadened. The frequency-broadened light, together with laser light scattered from static tissue, is photo-detected, and the resulting photocurrent is processed. The high spatial variability in blood flow across tissue such as the skin limits the clinical utility of single point LDF measurements. However, this limitation can be overcome with LDPI which provides a two-dimensional colour-coded representation of blood flow for the imaged tissue (figure 2(a)). Furthermore a key advantage of LDPI is that it is non-contact and the target tissue size can be configured to optimize the spatial resolution. Typical square field sizes range from a few centimetres up to 0.5 m (figure 2(b)).

Zoom In Zoom Out Reset image size

Briers (2001) provided an excellent topical review on LDPI technology and applications for blood perfusion mapping and imaging for the formative years of the technique. His review also described the theory for the measurement of blood flow changes in the microvasculature. Laser Doppler imaging technology is now available in several forms, using the traditional raster scanning method and more recently line scanning technology (Nguyen et al 2011) and full-field imaging capability (Serov and Lasser 2005, Leutenegger et al 2011, He et al 2012) (figure 3). The imaging systems usually use a single operating wavelength, i.e. red or near infrared. Murray and co-workers (2004, 2005, 2009a) have shown the clinical value of using a red wavelength LDPI system in conjunction with a green laser and this is discussed further below.

Zoom In Zoom Out Reset image size

The full-field fast scanning LDPI systems are relatively new to the market and so much of the evidence base for clinical applications lies with the raster line scanning systems.

LDPI has been applied extensively in clinical and research studies with reported applications in burn depth assessment, dermatology and plastic surgery, inflammation, wound healing, rheumatology, diabetes, pain, endothelial function assessment, cancer and angiogenesis. Fullerton et al (2002) provides guidelines on standardization in LDPI methods. Klonizakis et al (2011) address the reproducibility of microvascular measurements using laser Doppler techniques.

Burn wound depth assessment.

It is important clinically to determine if a burn is deep or superficial as this affects the choice of treatment. A deep burn usually requires grafting, whereas a superficial burn with skilled conservative management can heal on its own within 21 days without costly surgery. In deep tissue burns, the dermal layers and associated vasculature are damaged and so the hyperaemic blood flow response in normal wound healing is disrupted. A reliable hyperaemia will be present or absent after about 48 h post burn for superficial dermal or deep dermal burns, respectively, and use of LDPI to assess this (figure 4) has been validated, until 5 days post burn, by Monstrey et al (2011) and Pape et al (2012). LDPI can detect the microvascular blood flow in the dermis and, when used within this timeframe, can detect the presence of hyperaemia and thus give objective evidence for a burn being classed as deep or superficial (Hoeksema et al 2009, Pape et al 2012).

Zoom In Zoom Out Reset image size

LDPI is now a widely accepted tool to assess burn depth, with an ever-increasing body of evidence to support its use, as described in reviews (Devgan et al 2006, Sainsbury 2008, Monstrey et al 2008, Sharma et al 2011). The National Institute for Health and Clinical Excellence (NICE) (2011) provided evidence of clinical benefit and cost saving when LDPI (specifically for the Moor Instruments moorLDI2-BI imaging system) was used to guide treatment decisions for patients in whom there was uncertainty about burn depth and healing potential. This was a reliable endorsement to support both the clinical efficacy and health economics of the technology.

Dermatology and plastic surgery.

Surgery and wound healing.

Rheumatology.

Diabetes.

Endothelial (dys)function.

Endothelial function can be assessed from a macrovascular perspective e.g. by testing the flow-mediated dilatation response of the brachial artery during post-cuff occlusion reactive hyperaemia, whereby a reduced maximal vessel opening is associated with endothelial dysfunction. Endothelial function can also be studied by focusing directly on the behaviour of blood flow in the capillary vessels (Newton et al 2001b, Beer et al 2008). The capillaries in the skin are made from endothelium, and as they are close to the surface, they are accessible for measuring the degree of vasodilatation in their locality on provocation (Clough 1999). LDPI imaging is suggested to offer signal-to-noise advantages over conventional non-imaging laser Doppler flowmetry (LDF) as it allows averaging over many imaging points to reduce the effect of spatial heterogeneity.

There are a range of protocols for testing endothelial function using LDPI and these include local skin heating (water heater placed on the skin site, usually the forearm, and with an optical aperture for blood flow imaging, figure 5), iontophoresis (vasoactive compounds pass through the skin on a low level dc current stimulation, usually acetylcholine for endothelial-dependent vasodilatation (Ach) and sodium nitroprusside (SNP) for endothelial-independent vasodilatation as a reference comparator) (Morris et al 1995, Anderson et al 2004, Turner et al 2008), and post-occlusion reactive hyperaemia (flush characteristics assessed after usually a 5 min arm pressure cuff occlusion of at least 250 mmHg). Each of these methods provides different information relating to the health of the endothelium. Choice of measurement site and measurement reproducibility are very important considerations (Kubli et al 2000), with a coefficient of variation typically of 17% found for acetylcholine iontophoresis with LDPI (Newton et al 2001b). Normative data are important to establish for each measurement facility, protocol, local study population and age range (Elherik et al 2003, Millet et al 2012).

Zoom In Zoom Out Reset image size

2.1.3.1. Chronic pain.

Cancer and angiogenesis.

Brain and brain injury.

Chronic fatigue syndrome.

Briers (2001), Cheng et al (2004) and Basak et al (2012) describe the principle of LSCI. Laser speckle arises from the interference pattern created when laser light is scattered from an illuminated surface. The speckle contrast of the pattern, K, is defined as the ratio of the standard deviation (σ) to the mean intensity (〈I〉):

$$\begin{equation*} K = \frac{\sigma }{{\langle I \rangle }} \end{equation*}$$

which for scatter from a perfectly-diffusing (Gaussian) surface will be unity, but will be less than this in practice.

If there are moving scatterers associated with the illuminated surface (moving erythrocytes, for example), the speckles will become de-correlated over time and the contrast will vary. Integrating K over a short exposure time T (typically 20 ms) and across a small surface area (typically 5 × 5 pixels) allows an estimate of the speckle correlation time τ_c which will in turn be proportional to the velocity of the scatterers.

Figure 6 shows a theoretical plot of image contrast K against the ratio of correlation time to exposure time for a Lorentzian velocity distribution, whereby it is assumed the mean velocity is zero with an equal likelihood of positive and negative velocities. This would be a reasonable model to represent the random architecture of the microvasculature. Figure 6 also demonstrates how integration time T can be adjusted to select an appropriate range of erythrocyte velocities. It is necessary to operate along the steepest part of the curve to achieve the greatest sensitivity to change in perfusion.

Zoom In Zoom Out Reset image size

Recent improvements in processing power mean that computational time no longer limits the imaging rate, since it is shorter than the data acquisition time. Image resolution is reduced by the necessity for averaging over a 5 × 5 pixel window, but resolution in practice is normally limited by photon scattering under the tissue (Boas and Dunn 2010). Figure 7 shows a schematic configuration of a laser speckle contrast imager.

Zoom In Zoom Out Reset image size

The laser-speckle contrast technique has been widely validated in vivo against other established methods of microvascular perfusion measurement.

Bezemer et al (2010a) showed a good correlation between LSCI and sidestream dark field microscopy at the human nailfold during gradual occlusion of the upper forearm, and subsequent reperfusion. A number of studies have investigated the equivalence of LSCI and laser Doppler techniques. Binzoni et al (2013) reported poor agreement between LSCI and point laser Doppler at the human forearm. Differences in readings may arise in part because of different measurement depths achieved by the two techniques. Tew et al (2011) demonstrated better reproducibility of post-occlusive reactive hyperaemia at the forearm with LSCI than with laser Doppler flowmetry, probably due to the larger surface area sampled with LSCI. Conversely, sympathetic vasomotor reflexes at the finger pad (deep inspiratory gasp, cold pressor test) were more reproducible with laser Doppler flowmetry than with LSCI.

Millet et al (2011) compared LSCI with LDPI in the assessment of skin blood flow at the human forearm. LSCI showed lower inter-site variability of basal blood flow than LDPI, probably due to the deeper penetration of LDPI into the tissue, thus sampling the high-variability blood flow in forearm veins. There was a good correlation between the two techniques overall, but there was a proportional bias between the methods that limited their agreement.

Mahé et al (2012) have reviewed the use of LSCI for the assessment of skin microvascular function and dysfunction, commenting in particular on the high reproducibility of speckle contrast techniques in comparison to laser Doppler modalities.

LSCI has been applied extensively in human (and animal) studies of cutaneous, cerebral and renal blood flow.

Skin microvasculature.

Cerebral and stroke.

Renal.

Fukuoka et al (1999) used a laser-speckle technique during surgery to measure blood flow in the subchondral bone of the femoral head in 100 patients with osteonecrosis. LSCI was able to distinguish the ischaemic from normal tissue areas in 92 of the cases. LSCI is also of utility in the assessment of blood flow in articular tissues. Miller et al (2010) described the measurement of blood flow in the medial collateral ligament in a rabbit model of anterior cruciate ligament deficiency. Other potential applications include thyroid eye inflammatory disease in endocrinology. An example of this is shown in figure 8 for a patient with the active form of the disease. Determining if inflammatory activity around the eyes is significant gives valuable care pathway information. Active patients are treated with anti-inflammatory medication rather than corrective surgery (which should only be done when the patient is not active).

Zoom In Zoom Out Reset image size

All objects at a temperature above absolute zero emit energy according the Stefan–Boltzmann law (Ring et al 2009). If an opaque surface at absolute temperature T_abs is imaged against an absolute background temperature T_bgrd, the detected power Q is calculated from:

$$\begin{equation*} Q = \sigma \varepsilon AT_{\rm abs}^4 + \sigma ( {1 - \varepsilon } )AT_{\rm bgrd}^4 \end{equation*}$$

where σ is the Stefan–Boltzmann constant (5.67 × 10⁻⁸ Wm⁻² K⁻⁴), A is the surface area of the object, and ε is the emissivity of the material: a correction factor ranging from 0 to 1 that indicates how well an object emits radiant energy.

A theoretical perfect emitter would have an emissivity of unity: such an emitter is known as a blackbody. Human skin has been shown to behave very similarly to a blackbody (Sanchez-Marin et al 2009), making the human body an efficient infrared radiator.

The above expression calculates the power detected over all wavelengths, but a thermal imager will only be sensitive to infrared radiation across quite a narrow bandwidth, and as object temperature changes the amount of energy within the imager bandwidth will also change. Additionally, the response of the thermal detector and camera optics may not be flat over the bandwidth used. A more representative expression for the relationship between the temperature and the signal measured by a thermal camera detector would be:

$$\begin{equation*} {\rm Signal} = \varepsilon f(T_{\rm abs} ) + (1 - \varepsilon )f(T_{\rm bgrd} ) \end{equation*}$$

with the nature of the temperature functions (f) derived experimentally for the detector used, and optimized for the temperature measurement range required.

Figure 10 shows a schematic for a typical thermal imager design. Incident radiation is focussed by a filter and lens arrangement onto the surface of an infrared detector system. This may consist of a single detector which is mechanically scanned across the field of view by a system of rotating mirrors or, more commonly in modern devices, it may comprise a large collection of 'staring' infrared detectors fabricated onto a single microchip. In this case there is a one-to-one correspondence between each detector on the array and each pixel within the output thermal image. Such detectors are termed focal plane arrays (FPAs) because the readout circuitry is incorporated onto the detector chip to provide complete processing of the image in the focal plane.

Zoom In Zoom Out Reset image size

There is a wide variety of detectors available, which can operate in the 'far' (7.5–15 µm), 'mid' (2–5.6 µm) or 'near' (0.8–2 µm) thermal infrared wavebands. 'Photon' type detectors are doped semiconductors, with a current generated when infrared photons impinge on the detector and excite electrons across an energy gap. Such detectors require cooling to reduce noise (which complicates the imager design), but benefit from high thermal sensitivity. A less expensive solution is the uncooled microbolometer detector. Here, each detector in the array is heated by incident infrared radiation, causing a change in resistance. These detectors are typically of lower thermal sensitivity than their cooled counterparts, and are more prone to drift with change in ambient temperature. Nonetheless, refinements in design and fabrication have seen a radical improvement in uncooled detector performance in recent years.

Thermal imagers are generally not marketed as medical devices (sales are instead targeted at industrial and maintenance applications). This means that biomedical users of thermography must ensure procedures are in place in-house for rigorous quality assurance and calibration of the thermal imager (Howell and Smith 2009, Plassmann et al 2006). Objective assessment of performance also allows imagers to be compared, and the most appropriate device selected for clinical applications (Richards et al 2013).

The relation between skin temperature and microvascular dermal perfusion is complicated, and dependent on body site. Away from the periphery, skin temperature will be additionally strongly influenced by factors such as subcutaneous and muscle blood flow, skin thickness and perspiration. These in turn will be influenced by variables such as ambient temperature and core body temperature. A number of thermographic studies have investigated skin temperature across the body surface (Uematsu 1986, Zaproudina et al 2008, Goodman et al 1986). Howell et al (2009) compared temperature with laser Doppler blood perfusion at various skin sites.

In contrast, the acral areas play a key role in autonomic thermoregulation, and change in microcirculatory perfusion is the dominant driver of temperature change in glabrous (hairless) skin, which is richly populated with arteriovenous anastomoses. Figure 11 plots nose blood flow (monitored by laser Doppler flowmetry) against nose temperature as ambient temperature is varied across the range 15–40 °C. As ambient temperature rises above core body temperature, skin temperature is kept below 37 °C by a significant increase in skin blood flow. Seifalian et al (1994) gained a similar asymptotic relationship between skin blood flow and skin temperature by comparing laser Doppler perfusion with thermography measurements on hand skin.

Zoom In Zoom Out Reset image size

After cold challenge of the hand, Pauling et al (2012b) found a good correlation in healthy subjects between thermography and hand skin perfusion as measured by the laser speckle contrast technique, and the correlation was strongest in glabrous skin. Buick et al (2009) also found a similarity between thermography and LSCI when measuring response to hand cold challenge at the fingertips. However the two techniques showed a different response to a skin scratch at the volar surface of the forearm: the speckle instrument revealed an inflammatory reaction at the scratch site, whereas thermography predominantly highlighted increased flow in nearby perforator vessels bringing blood flow to the scratch (figure 12).

Zoom In Zoom Out Reset image size

Merla et al (2007) formulated a heat-flow model through the skin and were able to convert a thermal image into an image estimating cutaneous blood flow (in arbitrary units) at each pixel point. These cutaneous blood flow estimates correlated well at the dorsum of the hand with blood perfusion as measured by LDPI (laser wavelength 670 nm). No data was presented for the correlation between the two methods specifically at the glabrous skin of the fingertips.

2.3.4.1. Assessment of peripheral vasospasm.

Surgical applications.

Capillaroscopy is a non-invasive imaging technique that is used for in vivo assessment of the microcirculation (Bollinger and Fagrell 1990). It is a microscope technique that allows the morphology of capillary loops to be studied, typically with a 'wide field view' covering a width of a few tens of loops (a few millimetres). It is often applied to the nailfold area of the fingers (NFC) but depending on microscope design could be used to study other areas of the body, including the lips, tongue and mouth (Awan et al 2010, Scardina and Messina 2012, Scardina et al 2013). It is conceptually a simple technique but nevertheless it can provide valuable diagnostic information in the clinical microvascular setting. The nailfolds at the toes can also be studied, however it has been shown this does not provide equivalent diagnostic information to the finger site in patients with either connective tissue disease or vascular disease (Jung and Trautinger 2013).

A principal and validated clinical role is in the differential diagnosis of specific connective tissue diseases, for example SSc (Maricq and LeRoy 1973, Herrick 2008, Cutolo 2010). Other connective tissue diseases of the scleroderma spectrum, such as dermatomyositis, can also be assessed. As well as providing morphological information on capillary structure in health and disease, capillaroscopy can be used to obtain quantitative data such as red blood cell velocities at rest and in response to a stimulus, and also leakage fluorescence studies.

NFC can be performed with various optical instruments, from a standard clinical dermatoscope to a dedicated capillaroscopy system having tailored tissue lighting, a micro-positioner for accurate focusing and image capture and analysis functions. There are commercially-available monochrome and colour systems, some are portable skin contact devices and others non-contact, benchtop microscope configurations. A drop of immersion oil is added to the nailfold to aid visibility of the capillary loops. A cold illumination source is used, e.g. a high intensity LED. In a monochrome system this is often green or blue to give maximal contrast of the red blood cells. A white light source is employed for colour imaging systems. An NFC system needs the correct magnification lens to enable wide field or narrow field views, with these typically quoted as system magnification rather than true optical magnification e.g. x100 for skin width at approximately 3 mm displayed on a standard computer screen or monitor. Close-up lenses are typically x300. Once captured, images can be post-processed to enhance quality, quantitatively analysed to calculate loop size and density, printed in a patient report and archived (figure 13).

Zoom In Zoom Out Reset image size

NFC not only can help visualize the surface microvessels by microscopy, it can also be employed in combination with sophisticated methods in order to measure red blood cell velocity (figure 14), capillary pressure (i.e. cannulated capillaries using micropipettes and micropressure devices) (Morris et al 1996, Shore 2000) and transcapillary diffusion of a fluorescent tracer (sodium fluorescein injected intravenously—passage through vessels evaluated with a fluorescence microscope), thus allowing comprehensive physiological and pharmacological studies in humans (Bollinger and Fagrell 1990).

Zoom In Zoom Out Reset image size

The microcirculation has a fundamental role in the genesis of many autoimmune diseases including connective tissue disorders and diabetes. Before studying pathological changes, it is important to be able to identify normal capillary morphology and blood velocity patterns (Cutolo 2010). In healthy adults the capillaries usually maintain a constant overall morphology and so sequential follow-up can be applied to monitor for abnormalities. Children however can exhibit irregularity in their capillary morphology, and loop density has been shown to be lower than in adults. Adults of advancing age can progressively develop mild, non-specific morphological changes including tortuousity and micro-aneurysms (Dolezalova et al 2003).

Raynaud's phenomenon and connective tissue disease.

More than 10% of patients who attend a specialist centre with RP will eventually develop a scleroderma-spectrum connective tissue disease (Spencer-Green 1998). Along with the presence of specific autoantibodies, an abnormal NFC is a key risk-factor for the development of SSc in patients presenting with RP. SSc is a multi-organ disease characterized by tissue fibrosis and immune/microvascular abnormalities. The disease is very important to diagnose early so that it can be treated aggressively at an early stage (Cutolo et al 2008, 2010) (figure 15).

Zoom In Zoom Out Reset image size

The presence of giant capillaries and micro-haemorrhages on NFC is sufficient to identify the 'early' scleroderma pattern, and an increase in these features along with the progressive loss of capillaries (active pattern) is followed by neo-angiogenesis, fibrosis and 'desertification' (late pattern) (Cutolo et al 2013). The sensitivity of the American College of Rheumatology's classification criteria for SSc increases from 67% to 99% with the addition of these specific NFC abnormalities. Telangiectasiae may form on the face and hands along with digital ulcers. Based on the appearance of the scleroderma pattern on NFC, almost 15% of patients progress from primary to secondary RP over a mean follow-up period of 29 ±10 months. Follow-up by NFC (every six months) is suggested for RP patients. A scoring system for NFC changes is available, and scores change significantly during follow-up of SSc patients (Cutolo et al 2013).

Several other NFC patterns have also been identified, with non-specific microangiopathy that can be seen in patients with Sjogren's syndrome (SS), systemic lupus erythematosus (SLE) and undifferentiated connective tissue disease (UCTD). Significant microangiopathy is also often evident in dermatomyositis. It is important to note that mild changes from the normal range can be seen in some patients with primary RP and acrocyanosis.

Diabetes.

Non-specific microangiopathy at the nailfold site can also be identified in a variety of other conditions. Care needs to be taken in clinical capillaroscopy examinations to know a patient's wider health status when reporting on morphology. For example, Ribeiro et al (2012) showed that a group of 46 dermatology patients with psoriasis had a significantly lower capillary density, increased avascular areas and increased general morphological abnormalities compared to a group of 50 control subjects. However, no association was found between capillary density and the duration of the disease or the extent of skin involvement as measured by the psoriasis area and severity index (PASI) score. Furthermore, the presence of avascular areas was more common in psoriatic individuals whose nails were affected by the condition. NFC may also give valuable information about some features of patients with glaucoma. In a study by Park et al (2011) of 108 glaucoma patients, nail bed haemorrhage and loss of nailfold capillaries were strongly associated with the presence of optic disc haemorrhage. No differences were observed between patients with normal tension glaucoma and patients with primary open-angle glaucoma.

Skin ischemia is one of the crucial phenomena during chronic lower limb ischemia in patients with peripheral arterial occlusive disease and/or diabetes. However, risk stratification for development of ischemic ulceration and/or skin necrosis in those patients is not easy. Monitoring of microcirculatory parameters, as part of an integrated diagnostic approach, may have a considerable value in the evaluation of risk of progression of the disease and the effectiveness of therapeutic intervention in individual patients. Capillaroscopy has been used by many groups worldwide in relation to wound healing, including venous ulcer healing (Ambrózy et al 2013), critical limb ischaemia assessment (Kluz et al 2013), and its response to spinal cord electrical stimulation therapy (Colini Baldeschi and Carlizza 2011).

Capillaroscopy has also been applied to assessing endothelial function by measuring skin capillary density and recruitment and also red blood cell velocity changes with post-occlusive reactive hyperaemia (Cheng et al 2013). This approach has been reported in testing anti-hypertensive treatment (Kaiser et al 2013) where capillaroscopy was used to assess, in 44 hypertensive patients and 20 age- and sex-matched controls, if metoprolol succinate (a β1 -adrenoceptor blocker) or olmesartan medoxomil (an angiotensin II AT₁-receptor blocker) reverses microvascular dysfunction in hypertensive patients. They found that capillary rarefaction and microvascular endothelial dysfunction in the patient group responded favourably to long-term pharmacological treatment.

Traditional photoplethysmography (PPG) is a non-invasive contact optical method to detect the cardiovascular pulse wave, usually from the peripheral tissues (Allen 2007). The technology in its most basic form requires only two optoelectronic components: a light source to illuminate the tissue and a photodetector to measure small variations in light intensity arising from light interaction with the illuminated tissue. The origin of the changes in light intensity are not fully understood, although several mechanisms are thought to be involved including the shifting orientation of red blood cells between systole and diastole, vessel wall movement, and variations in tissue blood volume. PPG has been well reported in the literature and has a vast range of clinical applications including oxygen saturation (pulse oximetry), heart and respiration rates, blood pressure and cardiac output, assessment of autonomic function, and detection of peripheral vascular disease (Allen 2007). PPG is a tissue contact technique and is limited to a single or small number of measurement spots.

iPPG has been developed recently to give full-field and non-contact images of tissue pulsatility and perfusion. Measurement systems usually operate in reflection mode where both the illuminating light source and photodetector are situated alongside each other in dual or triple wavelength configurations. Typically, 660 nm and/or 880 nm ring illumination sources are used for dual wavelengths. Wieringa et al (2005) has described the development of SpO₂ imaging technology based on three wavelengths: 660, 810 and 940 nm. Sun et al (2012) have described the use of ambient light in remote photoplethysmographic systems to assess the utility of low-cost webcams to detect the pulse from the skin.

The detector technology usually comprises a high sensitivity CMOS camera which is held at a fixed distance, e.g. 10–20 cm from the tissue, to measure the variations in light (typically 2–5% of overall reflected light). Images can be collected at rates of tens of frames per second (fps), enabling the pulsatile components to be extracted without aliasing. Image processing comprises various stages, including band-pass filtering and averaging of regions of interest (ROI). The time-varying intensity modulation is strongly affected by relative movement of the camera and body so image processing methods are used to reduce the effect of motion artefacts e.g. by averaging of pixels or Fourier analysis techniques (Wieringa et al 2005, Verkruysse et al 2008). Other centres have experimented with lock-in amplifiers to improve the signal-to-noise ratio (Kamshilin et al 2011). The sensitivity of modern systems allows measurements to be made using very low cost webcams and distances from camera to measurement surface can now typically be 1 m (Rubins et al 2011) (figure 17).

Zoom In Zoom Out Reset image size

Early iPPG imaging systems were limited by low frame sampling rates, restricting their clinical use, for example in the assessment of pulse rate and its variability. Developments have been promising in terms of speed and sensitivity. The key physiological parameters i.e. respiration rate, heart and also pulse rate variability (PRV) derived from the iPPG datasets can yield statistically comparable results to those acquired using a contact PPG sensor (Sun et al 2013). Advanced physiological monitoring has been reported by Hu et al (2009, 2010) using opto-physiological modelling to extract information on superficial and deeper tissue blood volume perfusion levels (figure 18).

Zoom In Zoom Out Reset image size

iPPG technology development has advanced in the last few years, but with few clinical validation papers published in the literature. Early validation has been mainly in healthy controls and has included monitoring simple vascular responses using scratching of the skin (Kamshilin et al 2011), reactive hyperaemia following removal of a temporary vascular occlusion (Kamshilin et al 2013), and measurements made during exercise (Sun et al 2011).

Wieringa et al 2005 described the development of a method for imaging the arterial oxygen saturation (SpO₂) distribution within tissue. They used a three wavelength system with a monochrome CMOS camera, an apochromatic lens and a three lambda-LED ring light (i.e. 660, 810 and 940 nm) and compared images with respiration and ECG (heart rate) measurements in seven volunteers. Movies were processed by dividing each image frame into discrete ROI, averaging 10 × 10 raw pixels each. For each ROI, pulsatile variation over time was assigned to a matrix of ROI-pixel time traces with individual Fourier spectra. Images derived from photoplethysmograms correlated well with respiration reference traces at the three wavelengths. This early feasibility study also showed the potential for non-contact 2D imaging reflection-mode pulse oximetry.

Rubins et al (2011) evaluated the reliability of a high resolution webcam based single wavelength system by comparing its measurements with LDPI at the palmar surface of the hand during local warming. Results showed that the amplitude of the iPPG increased immediately after warming and had good correlation with the mean LDPI perfusion (correlation 0.92, p <0.0001), giving confidence that the iPPG technique has good potential for the non-contact monitoring of blood perfusion changes.

Sun and co-workers (2011) assessed the performance of an iPPG system by measuring the cardiac pulse from images of the hand in 12 subjects before and after 5 min of cycling exercise. A new motion artefact reduction method was implemented based on planar motion compensation and blind source separation. The physiological parameters (i.e., heart rate, respiration rate) derived from the images captured by the iPPG system exhibited functional characteristics comparable to conventional contact PPG sensors. Such continuous recordings from iPPG reveal that it is possible for heart and respiration rates to be successfully tracked in high-intensity physical exercise situations. Although monitoring pulse rate has been shown to be feasible by iPPG, the precise measurement of heart rate variability is challenging. Sun et al (2013) also described a system capturing iPPG images at 200 fps. An important outcome was that the negative influence of a low sample frequency can be compensated via interpolation to improve the time domain resolution. This work therefore showed the potential of low-cost webcam-based iPPG for the evaluation of cardiac and peripheral autonomic activity.

Kamshilin et al (2013) demonstrated a novel method of iPPG image processing for tracking blood flow changes in skin before, during and after arm occlusion. They showed the extensive variability in the 2D distribution of blood pulsation between cardiac cycles and spatially, as would have been seen if measured using another established microvascular perfusion imaging technique such as LDPI.

The technique employs a low power laser, usually infrared, shone onto tissue. Boundaries between the tissues reflect back a small amount of light, and this reflected signal is detected and processed. Image construction arises primarily from the refractive index variations at these tissue discontinuities and micro-structures. OCT with a near infrared light source can give information about the epidermis-dermis junction in skin.

OCT is based on interferometry i.e. the interference between reflected light and a reference beam, which is used as a coherence gate to isolate light from specific depths. A Michelson interferometer is often used to perform the low-coherence interferometry for OCT (Fujimoto et al 2000). The frame rate for early OCT systems was typically 4 to 8 fps but faster modern systems are available at beyond video frame rate (depending on imaging parameters). A schematic representation of an OCT system is shown in figure 19 and example OCT B-mode image in figure 20.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

There are various OCT systems available. Michelson Diagnostics (Orpington, Kent UK) is an established manufacturer of clinical OCT technology. Their 'VivoSight' optical OCT probe comprises four parallel swept source OCT systems using a laser with central wavelength of 1310 nm and with a 150 nm sweep. It can be configured to automatically capture 'N' brightness (B-) mode scans with a set interspacing to produce a 3D re-construction of the tissue, a typical image size obtained being 4 × 0.4 × 2 mm. Axial (A-) line scanning information is also available.

Second-generation OCT systems, known as Fourier (frequency) domain OCT, are now available including the Thorlabs Spectral Radar OCT imaging system (Newton, NJ, USA). Frequency domain techniques have several advantages over time domain systems including a higher sensitivity and speed. The improved speed allows for rapid cross sectional image collection, enabling facile imaging of larger sample volumes in real-time. Additional contrast mechanisms are able to extend the capability of OCT. Doppler OCT or optical Doppler tomography is one kind of functional extension of OCT which combines the Doppler principle with OCT and provides in-vivo functional imaging of moving samples, flows and moving constituents in biological tissues (Liu 2012) (figure 21).

Zoom In Zoom Out Reset image size

The selection of wavelength for a particular OCT system depends on specific application requirements. Image quality depends both on the type of sample, as well as system design. For example, water is more transparent to light in the 600–900 nm wavelength range. Since the outer portion of the eye (cornea, vitreous humour and lens) is mostly water, OCT imaging at 800 nm has become the industry standard in many ophthalmic applications. For multi-layered biological tissue (skin, brain, GI tract, etc) the 900–1400 nm range is appropriate, since longer wavelengths typically provide better penetration depth.

In general, the resolution depends on the coherence length of the light source. Shorter coherence lengths, which are associated with broader spectral bandwidths and shorter wavelengths, lead to better axial (longitudinal) image resolution. For most OCT applications the axial resolution is approximately 10 µm. With a measurement depth of the order of 1 mm this should allow imaging of the stratum corneum, the epidermis with the basal membrane zone, and the structures of the upper dermis (which consists of fibroblasts embedded in a network of collagen fibres and small blood vessels).

Although OCT is becoming established in specific clinical specialties including ophthalmology (Hahn et al 2011), histology (Jung and Boppart 2013), dentistry and human oral cavity imaging (Davoudi et al 2012), and cardiology, its application appears focussed to the study of the microstructures associated with the small blood vessels rather than specifically the blood vessels per se.

Cardiology OCT has been used in coronary artery assessment (Nammas et al 2013) giving a near tenfold improvement in resolution over intravascular ultrasound. Valuable information was extracted for the in vivo study of vascular scaffolds in stents around a coronary artery, and vessel wall injuries such as dissection, thrombosis and tissue protrusion. Of course, coronary arteries are significantly larger than the capillary vessels but the studies show the potential for studying microangiopathy. A review of OCT applications in dermatology by Gambichler et al (2005) showed the potential for 'histological' assessment of the superficial skin layers, skin appendages, blood vessels, cutaneous inflammation, hyperkeratotic conditions and photoadaptive responses.

An example OCT image of the nailfold skin in a patient with SSc is shown in figure 22. The cross-sectional view indicates the region of dilated blood vessels. This imaging shows the fine static detail of the tissues and a further advancement of the technique would be to superimpose Doppler OCT blood flow information onto the image (Mason et al 2004).

Zoom In Zoom Out Reset image size

Doppler OCT holds great promise for ophthalmology applications. Willerslev et al (2013) used spectral domain OCT and showed that the direction of blood flow at dichotomous vascular branchings can be determined. This feature may assist in the identification of flow reversal near sites of vascular occlusion. They also analysed blood flow near vascular malformations. Kagemann et al (2009) used a spectral domain OCT system to study a glass capillary tube of sub-millimetre diameter perfused with milk. They calculated the Doppler shifts from temporal changes in phase and showed that the observed percentages of the velocity profile at or below the Nyquist frequency was highly correlated with the predicted percentages. Cito et al (2012) successfully tested spectral domain OCT methods in various liquids passing through the microchannels of micro-fluidic devices.

An overview of the applications for OCT in imaging microvascular networks was given by Mahmud et al (2013) who described its potential roles for assessing retinal and choroidal vasculature, cardiac vasculature, central nervous system, and tumour models. They compared two general classes of microvascular imaging techniques and their algorithms: speckle-variance and phase-variance OCT.

Fujimoto and Drexler (2008) also review OCT technology and describe its use in imaging ocular blood flow, mapping cortical haemodynamics for brain research, drug screening, monitoring changes in image tissue morphology and haemodynamics following pharmacological intervention and photodynamic therapy, evaluating the efficacy of laser treatment in PWS patients, assessing the depth of burn wounds, imaging tumour microenvironment, and quantifying cerebral blood flow.

PT involves shining a very short light pulse (typically a few tens of ns width) onto tissue via an optical fibre. Wherever this light is absorbed, the energy from the light warms that region of tissue or blood, in the region of one thousandth of a degree Kelvin. This warming causes the absorbing region to expand rapidly by a very small amount, with this thermoelastic expansion causing a pressure wave to be set up which travels through the tissue at the speed of sound, before being picked up by a high frequency (e.g. 30–70 MHz) focussed ultrasound detector. The more a section of tissue or region of blood absorbs, the hotter it becomes and the greater the ultrasound signal it creates. So unlike a standard ultrasound image, the photoacoustic image reveals the different optical absorption properties of tissue or blood (figure 23). As with diffuse optical tomography different wavelengths of light, for example green and infrared, are used to reveal varying levels of oxygenation in the blood as well as different tissue types (extract from 'IPEM Spotlight on Optical techniques in Medical Imaging', 2006).

Zoom In Zoom Out Reset image size

The technique can provide 3D images with significant depth of penetration up to 7 cm. The image resolution is in the order of 10–100 µm. Image acquisition time can be of the order of tens of minutes. PT imaging has the important advantage that the photoacoustic signals can be detected on the same side and over the same region of tissue that is irradiated with the excitation light (figure 24). This is particularly important for imaging skin and superficial blood vessels (figure 25).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Wang (2008) summarized the great potential for the technology, including melanoma detection, endoscopy, molecular imaging, gene expression, Doppler PT for flow measurement, metabolic rate measurements, mapping of sentinel lymph nodes, and breast and brain imaging. PT can be combined with other imaging modalities, including OCT, to enhance the information derived in microscopy (Li et al 2009), and also endoscopy. (Xi et al 2013). Xiang et al (2013) have advanced the 3D imaging and developed a 4D PT technique that generates motion pictures of imaged tissue, enabling real-time tracking of dynamic physiological and pathological processes at hundred micrometer-millisecond resolutions. The 4D PT technique was used to image needle-based drug delivery and study pharmacokinetics. It has also been suggested as a technique to monitor the fast haemodynamic changes during interictal periods in epilepsy and also the temperature variations during tumour thermal therapy.

Specific clinical studies include the identification of inflammatory arthritis and also breast cancer. Rajian et al (2012) used PT to identify neovascularity (angiogenesis) which is an early feature of inflammatory arthritis, thus giving an early diagnosis and facilitating treatment monitoring of the disease. A rat model was tested with imaging at two different wavelengths, 1064 and 532 nm, to reveal that there is a significant signal enhancement in the ankle joints of arthritis-affected rats when compared to a normal control group. Histology images obtained from both the normal and the arthritis-affected rats correlated well with the PT findings, supporting the view that PT could become a new tool for clinical management of inflammatory arthritis.

Manohar et al (2005, 2007) explored the utility of PT in breast cancer patients. Near-infrared photoacoustic images (figure 26) of ROI in four of the five cases of patients with symptomatic breasts revealed higher intensity regions attributed to vascular distribution associated with cancer. Of the two cases presented, one was especially significant, with benign indicators dominating in conventional radiological images, while photoacoustic images revealed vascular features suggestive of malignancy, corroborated by histopathology. The results showed that PT may have potential in visualizing certain breast cancers based on intrinsic optical absorption contrast. A future role for the approach could be in supplementing conventional breast imaging to assist detection and/or diagnosis. Improvements in contrast have been shown in a mouse tumour model by combining high-resolution near-infrared light-induced PT with NIR dye-labelled amino-terminal fragments of urokinase plasminogen activator receptor (uPAR) targeted magnetic iron oxide nanoparticles (NIR830-ATF-IONP) (Xi et al 2014). The accumulation of the targeted nanoparticles in the tumour led to photoacoustic contrast enhancement due to the high absorption of iron oxide nanoparticles (IONP). Near infrared fluorescence images were used to validate specific delivery of NIR830-ATF-IONP to mouse mammary tumours. Systemic delivery of the targeted IONP to the tumours produced a four-fold enhancement in photoacoustic signals compared to tumours that received non-targeted IONP. The use of targeted nanoparticles allowed imaging of tumours located deeper than 3 cm beneath healthy tissues. The study indicated that the combination of PT and receptor-targeted NIR830-ATF-IONP can provide improved specificity and sensitivity for breast cancer detection.

Zoom In Zoom Out Reset image size

HSI is based on a spectroscopic technique and involves the generation of a 'hyperspectral' cube stack of spatial and spectral information. This '3D' data cube set comprises, at each pixel of the image, a full spectrum. The spectra generated would depend on both target illumination and detector characteristics. In practice, the field of view is externally illuminated with a broad spectrum light source, with uniform 'white light' illumination across the visible wavelengths (400–700 nm). The illumination can also extend to near infrared depending on what tissue is being studied.

A key medical application of HSI is the spatial non-invasive measurement of tissue oxygen saturation (SO₂). Although there are spot measurement devices commercially available that can measure SO₂ in skin, these are contact based and localized. The contact pressure can add to the uncertainty in the values measured. Furthermore, there are clinical applications where sterile non-contact methods are required such as in burns assessment and wound healing. HSI can calculate SO₂ from established methods by comparing absorption spectra in the green part of the spectrum where high oxygen saturation in tissue haemoglobin has a pulsatile spectrum whereas reduced haemoglobin is monophasic and damped (figure 27).

Zoom In Zoom Out Reset image size

A key advancement in HSI technology is the development of narrow band tuneable filters which have the characteristic ability to transmit electromagnetic radiation through a very narrow bandpass of spectral region over a wide spectral range (Yuen and Richardson 2010). Filter types include acousto-optical filters and liquid crystal tuneable filters, with each having specific advantages and disadvantages (Gat 2000). Challenges still remain with HSI development in dealing with mixed spectra and also the scattering effects in biological tissues.

There are many potential applications for HSI in assessing the microcirculation and associated perfusion, but the literature is currently limited and there are few clinical studies reported on the human microcirculation. Example HSI images are shown for SO₂ mapping in forearm skin sensitivity testing (figure 28) and in diabetic foot ulcer assessment (figure 29).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Nishino et al (2013) developed a method for the assessment of allergic dermatitis by using the long-wavelength near-infrared spectrum to detect intracutaneous allergic type-specific elements. A spectral canonical discriminant analysis (CDA) classifier was established for the spectra of type I and type IV allergic dermatitis reactions and achieved accurate detections by good specificity and sensitivities overall, suggesting a possible application of near-infrared spectral imaging to the assessment of allergic dermatitis.

Foot ulceration remains a major health problem for diabetic patients and has a major impact on the cost of diabetes treatment. HSI has great potential to quantify cutaneous tissue haemoglobin oxygenation in this group from generated anatomically-relevant tissue oxygenation maps to assess the healing potential of diabetic foot ulcers (figure 29).

Yudovsky et al (2011) developed an HSI algorithm based on an experimentally validated skin optical model and inverse algorithm to monitor temporal changes in epidermal thickness and oxyhaemoglobin concentration in diabetic feet at pre-ulcerative sites. They pilot tested this in a few patients. The algorithm was also able to observe reduction in the epidermal thickness combined with an increase in oxyhaemoglobin concentration around the ulcer as it healed and closed, showing potential for the methodology to be used for the early prediction of diabetic foot ulceration in a clinical setting.

Nouvong et al (2009) reported a larger study of a prospective single-arm blinded design in 66 patients with type 1 and 2 diabetes and followed over a 24 week period. Transcutaneous oxygen tension (tcPO₂) was measured at the ankles as a reference. Superficial tissue oxy-haemoglobin (oxy) and de-oxyhaemoglobin (deoxy) were measured with HSI from intact tissue bordering the ulcer and a healing index derived from oxy and deoxy values for assessing the healing potential. A total of 73 ulcers were studied in total, of which 19 did not heal. The sensitivity for the healing index was 80%, specificity 74%, and positive predictive value 90%. The results indicated that cutaneous tissue oxygenation correlates with wound healing in diabetic patients. They concluded that HSI of tissue oxy and deoxy may predict the healing of diabetic foot ulcers with high sensitivity and specificity based on information obtained from a single visit.

Ophthalmology and the study of the retinal blood vessels is another important area of opportunity for HSI. In a study by Shahidi et al (2013) they used an HSI system to investigate regional differences in oxygen saturation of blood in first degree retinal vessels using a novel non-flash hyperspectral retinal camera in nine healthy individuals. Specific images were collected at 548, 569, 586, 600, 605 and 610 nm. Optical density values were extracted with the aid of Image-J software for blood oxygen saturation (SO₂) determination. Arteriolar and venular SO₂ were measured at three locations (ranging 1–3 optic nerve head radii) from the disc margin along the vessels in the superior and inferior temporal quadrants. They found greater SO₂ values in the superior temporal first degree retinal arterioles and venules in young healthy individuals than in the equivalent inferior vessels. However, there were no detectable differences in retinal SO₂ along each of the major vessels, a finding that is consistent with the concept of these vessels not contributing primarily to gas exchange. Moreover, SO₂ was consistently higher in the arterioles than in the equivalent venules (p < 0.0001).

The optical properties of red blood cells in the reticular dermis are highly dependent on wavelength, with strong absorption at green wavelengths, but only weak absorption of red spectral components. Absorption by the static tissue in the dermis is, in contrast, almost invariant across the visible light waveband. Spectroscopic evaluation of the proportional absorption of red and green wavelengths from skin, flash-illuminated by a xenon lamp, thus allows an estimation of red blood cell concentration within the tissue.

To ensure the light detected by TiVi originates from the reticular dermis, a cross-polarization technique is employed to filter the light reflected from the epidermis. Linearly-polarized light incident on skin will be partly reflected by the epidermal surface, mostly retaining its polarization in the process. Another portion of the incident light will instead undergo multiple scattering events in the reticular dermis and become depolarized. Figure 30 shows the configuration of a TiVi imager. A polarizing filter is placed over the xenon light source, and a second orthogonal polarizer is positioned over the detector to reject the light reflected directly from the skin surface. Depolarized light scattered within the reticular dermis is passed to the detector.

Zoom In Zoom Out Reset image size

O'Doherty et al (2007) derived an algorithm for calculating a TiVi_index representative of red blood cell concentration in the reticular dermis within the physiological range (0–4% RBC fraction of tissue volume), which accounts for light absorption in the epidermis and is almost independent of incident light intensity and oxygen saturation within the physiological range of 70 – 100% O₂ saturation. McNamara et al (2010) express this algorithm in terms of a 1024 × 768 pixel TiVi output image matrix M_out, whereby:

$$\begin{equation*} M _{\rm out} = \frac{ M _{\rm red} - M _{\rm green} }{M _{\rm red} }\, {\rm e}^{ - p( (M_{\rm red} - M_{\rm green}) / M_{\rm red})} \end{equation*}$$

M_red and M_green are the red and green colour planes of the image respectively, and p is a factor adjusted to derive the best linear fit between M_out and red blood cell concentration. Monte Carlo modelling suggests a measurement depth for TiVi of around 400–500 µm in skin with typical optical properties i.e. well within the reticular dermis.

O'Doherty et al (2011) validated a high-speed TiVi system using an experimental model with red food dye pumped through tubing at pulsation rates similar to the cardiac cycle. Analysis of the images generated a power spectrum, with the peak power corresponding to the pumping frequency, along with harmonics. When the system was used to image red blood cell concentration at the human forearm and finger nailfold, peaks in the power spectral density were detected corresponding to the heart rate, plus a component attributed to vasomotion at 0.1 Hz.

Nilsson et al (2009) assessed the reproducibility and stability across four TiVi imagers when measuring from a plane surface of uniform colour. The maximum drift in output observed in any of the imagers on repeated measurements was only 1.02%. The mean TiVi output for all four imagers was also calculated, and the largest deviation from this mean observed in any individual device was found to be 4.14%. Dependence on image distance and size, and the influence of ambient lighting levels and surface curvature were evaluated. These factors did not unduly influence imager output provided highly curved surfaces were avoided.

O'Doherty et al (2009) compared TiVi, laser speckle contrast and laser Doppler line scanning instruments in measuring skin responses to the application of methylnicotinate (a vasodilator), a vasoconstricting corticosteroid, and forearm occlusion. TiVi exhibited a different response from the laser-based imaging techniques because of its sensitivity to red blood cell concentration rather than perfusion. The TiVi signal increased with forearm occlusion at 90 mmHg due to pooling of erythrocytes in the skin, but the effect was less pronounced with occlusion at 130 mmHg due to the rapid cessation of arterial inflow. TiVi was able to detect the vasoconstrictive effect of the topical corticosteroid, whereas the laser Doppler and speckle contrast devices were insensitive to vasoconstriction. TiVi and LSCI provided less deep measurements than the laser Doppler instrument, which could not detect the vasodilatation caused by methylnicotinate in the superficial skin. Petersen et al (2010) have also reported the evidence that TiVi may be more sensitive to vasoconstrictive pharmacological stimuli than laser Doppler methods, whereas the laser-based techniques may be more suited than TiVi for recording drug-induced vasodilatation.

Farnebo et al (2010) found a strong correlation between red blood cell concentration as measured by TiVi and laser Doppler perfusion measurements in assessment of post occlusive reactive hyperaemia at the forearm. The maximum responses and recovery times were found to be more reproducible when assessed by TiVi than by laser Doppler, probably because TiVi allows an evaluation averaged across an entire region of interest in the image, whereas laser Doppler data arises from a single point on the skin surface.

Zhai et al (2009) compared TiVi with colorimetry for the assessment of corticosteroid-induced skin blanching. They found that the two techniques had similar relative uncertainties, but the TiVi technique was not subject to the operator-dependent factors encountered in colorimetry such as variability in applied probe pressure or angle. TiVi was also able to detect blanching in cases of paradoxical skin erythema after corticosteroid administration.