1. Introduction
Photoacoustic imaging (PAI) holds strong potential in providing structural, functional and molecular information on tissue, with scalable resolution and imaging depth [
1]. Since the optical scattering is high in biological tissue, ballistic optical microscopic techniques cannot provide any useful information beyond a depth of 1 mm. PAI overcomes this difficulty since it involves acoustic detection and sound scattering in tissue is orders of magnitude lower than that of light. In PAI, short-pulsed light is irradiated on the tissue, and endogenous optical absorbers in the tissue absorb light resulting in a temperature rise [
2]. This transient temperature rise results in thermoelastic expansion and produces light-induced ultrasound (US) waves, which then can be detected by US detectors placed on the skin surface for reconstructing an optical absorption map with acoustic resolution [
3]. In a clinical context, PAI is easy to combine with US imaging and is capable of providing anatomical, functional, molecular, and metabolic information by utilizing the signature optical absorption contrast of the vasculature, hemodynamics, oxygen metabolism, biomarkers, and gene expression. Utilizing the useful information provided by PAI, a plethora of clinical applications have been explored in vascular biology [
4,
5,
6,
7], oncology [
8,
9], neurology [
2,
10,
11], ophthalmology [
12,
13], dermatology [
14,
15], gastroenterology [
16,
17,
18,
19,
20], osteology [
21,
22,
23,
24], and cardiology [
25,
26].
In laser-based PAI, where the tissue of interest is illuminated by a pulsed laser beam, the optical energy used usually ranges from tens to hundreds of mJ per pulse, with a typical pulse duration of 5–10 ns. Most of the commercial and research lab-made PAI systems utilize Q-switched Nd:YAG pumped OPO (optical parametric oscillator), Ti: Sapphire or dye laser systems. However, because of their high cost, larger footprint and strict requirement for eye-safety goggles and laser-safe rooms, these laser sources are not suitable for a clinical environment. Furthermore, the repetition rate of most high-power laser sources is relatively low (~10 Hz), which limits the imaging speed, especially when the signal-to-noise ratio (SNR) is not sufficient and frame averaging is a necessity. In recent years, laser diodes (LD) and light emitting diodes (LEDs) have been heavily explored to be used as an illumination source in PAI, resulting in portable, affordable and clinically translatable PAI systems [
6,
7,
27,
28,
29,
30,
31,
32,
33,
34,
35,
36,
37]. LD offers a higher pulse repetition rate (PRR) (typical 2–4 KHz), average power around 6 W, and optical energy of around 0.56–2.5 mJ per pulse, but is only available at wavelengths greater than 750 nm [
38,
39]. Additionally, for an LD-based PAI system, laser-safe rooms and goggles are requirements, just as in the case of conventional laser sources. On the other hand, LEDs, which are available in a wide wavelength range (e.g., 470, 520, 620, 660, 690, 750, 820, 850, 940 and 980 nm) provide lower optical energy in the range of µJ per pulse, but at a higher repetition rate (~16 KHz) offering the possibility to average more frames without compromising on temporal resolution. Compared to fixed pulse widths in lasers, the optical pulse width of an LED/LD source can be tuned based on the required spatial resolution and imaging depth [
39,
40].
The pulse width of LED/LD sources is tens of nanoseconds, whereas that of solid-state lasers could be less than ten nanoseconds. The temporal pulse width imposes a limit to the spatial resolution of the imaging system. For example, the 35-ns pulse width of the 850-nm LED corresponds to a spatial resolution of 52.5 µm (=35 ns × 1500 µm/µs). The LD usually offers more energy than LEDs. An LD light source’s (Quantel, Bozeman, MT) pulse width, for example, can be tuned from 30 ns to 200 ns, with the pulse energy correspondingly changing from 1–4 mJ. As a limitation, the LED array can only reach up to 0.200 mJ per pulse (highest reported optical output for 850 nm LED arrays). LD, even though it can offer higher pulse energy, is the same as class-IV lasers in terms of optical coherence and subsequent eye/skin safety issues. Owing to its portability, affordability, imaging speed and safety aspects, LED-based PAI holds potential in real-time functional and structural characterization of tissue in various superficial and sub-surface imaging applications and also to accelerate the clinical translation of PAI. Typically used laser, LD and LED performance are listed in
Table 1.
In this paper, we review the progress of LED-based PAI technology and its potential preclinical and clinical applications.
2. Fundamental Development of LED-Based PAI Technology
PAI has already demonstrated its unparalleled potential in multiple preclinical and clinical applications and is quite mature in a research setting. At this point, this technology is facing an exciting transition from bench to bedside and LD- and LED-based systems are being explored heavily because of their portability, affordability, and ease of use in a clinical setting. LDs operating in pulsed mode have been investigated by different research groups for multiple point-of-care applications [
36,
37,
38]. However, typical commercial pulsed LDs are available only in the near-infrared wavelength range, and combining multiple wavelengths in a handheld setting is a cumbersome process [
41,
42]. On the other hand, the LEDs could be fabricated to operate in a 400 nm to 1000 nm wavelength range (not continuously) with reasonable optical energy by developing arrays of multiple elements and overdriving them [
43]. Within this wavelength range, PAI could provide high contrast for melanin, hemoglobin, and fat to an extent, making LEDs one of the ideal illumination sources for multispectral PAI of tissue up to a depth of 1–1.5 cm.
An LED is a semiconductor device based on a p–n junction diode. A p-n junction diode is a two-terminal semiconductor device, which allows the electric current in only one direction and blocks the electric current in the opposite or reverse direction. If the diode is forward biased by applying a voltage, it allows the electric current flow. A small increase in voltage results in a significant change in current flow. Holes (from the P-type material) and electrons (from the N-type material) flowing across the junction promote strong electron–hole radiative recombination, resulting in the emission of a large number of photons [
44]. Typically, LEDs are designed for continuous wave (CW) operation, but it is also feasible to drive them with pulsed current. In the pulsed mode, output energy of LEDs is dependent on the peak current, and this can be far higher than the CW rated current, especially if the duty cycle is kept low (<0.1%) to avoid any thermal damage. Since LEDs, when overdriven in a pulsed mode, can generate significantly higher optical output than in conventional CW operation, these types of overdriven pulsed-mode LEDs are often referred to as high-power LEDs. However, operating an LED at excessively high drive currents may lead to faster ageing due to heat generation, and this can even cause immediate failure of the LEDs. The quantum efficiency of the device will also drop with increasing current [
45]. Considering this, it is important to design and develop efficient, safe electronic drivers and heat sinks to use LEDs in high-power mode.
About a decade ago, Hansen first proposed the use of LEDs working at a 627-nm wavelength as an inexpensive and compact excitation source for biomedical PAI [
46]. In this proof-of-concept work, he demonstrated the feasibility of using LEDs as a light source in PAI for the first time. The basic idea of creating pulsed high-power LED is that, when the low-power LED is overdriven, they can deliver optical output that is far higher than their normal specifications in CW-operation [
47,
48,
49,
50]. The LED they used was Luxeon LXHL_PD09 which has been measured to yield approximately 250 mW of light output when supplied with 1 A DC current. A derivative of the MOSFET-based circuit presented by Alton and Raji was employed as a driver for generating pulsed current [
51]. When the LED was supplied with 60-ns current pulses with peak value of 40 A, it was able to provide pulse energy of 400 nJ per pulse with pulse width of 60 ns, and light focusing was performed to generate the radiant energy required for generating a photoacoustic (PA) response. Also, 50,000 A-line signals were averaged to detect PA response from a non-realistic phantom. Based on the PRR of the proposed LED (200 Hz), this system requires 250 s to acquire an image, which was not good enough for imaging tissue in real-time. Owing to the advances in solid-state device technology and efforts of different research groups in the last decade, there were significant improvements in the performance of LEDs (improvement of pulse energy, PRR, etc.), which consequently resulted in the step-by-step development and commercialization of an LED-based PAI system that is comparable to a laser-based machine.
In 2013, Allen et al. proposed the use of high-power LEDs (CBT−120 from Luminus) working at 400-nm to 65-nm wavelengths as an illumination source in biomedical PAI [
43]. In this work, the pulse energy of the LED is increased to 22 µJ per pulse with a pulse width of 500 ns, by overdriving the LED elements by 10 times their rated current. The driver they used was made based on MOSFET, which is described in work of Chaney et al. [
51]. As a result of increase in light energy, with the same PRR of 200 Hz, they were able to significantly reduce the frame averaging (by 1000 times), which is commendable.
In 2016, Allen et al. further improved their system performance by using high-power LED (SST−90 from Luminus) elements working at 400-nm to 650-nm wavelengths, driven by a commercial electronic driver (PCO−7120, Directed Energy, Inc., Loveland, CO, USA), which provided 9 µJ per pulse, with a pulse duration of 200 ns, a peak current of 50 A and a PRR of 500 Hz when overdriving LEDs by 20 times their nominal current [
33]. They confirmed that the duty cycle was 0.01%, still below the 1% which has been previously reported as safe (no noticeable damage to the device) [
45]. In this work, they first imaged a realistic tissue-mimicking phantom by averaging 5000 image frames and using a wide field illumination strategy. The best imaging depth they achieved was 15 mm in 1% intralipid (
Figure 1).
The pioneering works mentioned above laid the foundation for LED-based PAI, especially demonstrating the feasibility of using visible light for PAI, which is not possible using lasers or LDs. Oxygen saturation imaging is one of the most important applications of PAI, and it is of paramount importance to have illumination wavelengths suitable for this. Considering the absorption peak for deoxy-hemoglobin (690 nm) and oxy-hemoglobin (850 nm), it was critical to develop LEDs in these wavelength ranges too. Considering this, there has been significant efforts from different researchers in this direction, which are detailed below.
In 2016 and 2017, Agano et al. for the first time proposed the use of high-power LED arrays working at 850 nm with optical energy of 200 µJ per pulse and with a pulse duration of ~70–100 ns. They optimized the SNR by increasing the optical output, by developing highly noise-efficient front end electronics with multiple stages of amplification (~106 dB), and also by matching the US transducer’s frequency characteristics with LED light pulse width [
34,
35,
52]. To increase the optical output, they developed LED arrays and further optimized them using hybrid techniques. The single LED element provided output energy of 0.024 µJ per pulse, with a pulse duration of 70 ns and 1 A DC-current. By developing LED elements with double stack structure, arranging them in an array, and applying 20 times the rated current, they were able to achieve 200 µJ per pulse at a wavelength of 850 nm. The repetition rate of this first commercial LED-based PAI system (AcousticX) was 4 KHz. In 2018, Zhu et al. improved the PRR of LEDs to 16 KHz and used the same system for demonstrating its dynamic structural and functional imaging capabilities [
6]. The performance of the mentioned LED-based PAI technology is summarized in
Table 2.
The AcousticX system was thoroughly characterized for its imaging depth, spatial resolution, frame rate, and oxygen saturation imaging accuracy in multiple studies. When using a 9-MHz US probe and 850-nm LED arrays, Xia et al. reported mean axial and lateral resolutions of 220 µm and 460 µm respectively, which was similar for both US and PA imaging. They also reported an imaging depth of 2.8 cm at an interleaved US and PA frame rate of 30 Hz. Imaging depth was further improved to 3.8 cm after averaging more frames, resulting in a final frame rate of 1.5 Hz. In another study, Hariri et al. reported mean axial and lateral resolution of 268 µm and 570 µm respectively, when using a combination of a 7-MHz US probe and 850-nm LED arrays. They also obtained an imaging depth of 3.2 cm (frame rate = 15 Hz) in their phantom study. In this molecular imaging study, they also measured the sensitivity of the system in detecting commonly used molecular contrast agents. The limits of detection for ICG, MB, and DiR were reported to be 9 μM, 0.75 mM, and 68 μM, respectively. Oxygen saturation imaging is one of the most important applications of PAI. Using a two-wavelength (750/850 nm) approach, Kalloor Joseph et al. evaluated the potential of AcousticX in oxygen saturation imaging using phantom and in-vivo small animal imaging experiments. Based on 28 human-blood-based in-vitro measurements, a standard error of 8.4% was observed between actual oxygen saturation values and the oxygen saturation image formed by the system. In the same work, they showed repeatability and reproducibility of oxygen saturation imaging using an in-vivo mouse oxygen breathing challenge experiment.
From the beginning of 2017, there has been a tremendous push in this area and multiple studies using LED-based PAI were reported. In the next section, we will review the potential preclinical and clinical applications demonstrated using AcousticX, the commercially available LED-based PA and US imaging system.
4. Discussion
In this paper, we first reviewed the historical development of LED-based PAI, starting from the first report on single-point measurements to latest clinical pilot studies using high-power LED arrays. In the span of 10 years, there has been significant growth in this field, especially with the improvement of pulse energy (nJ to hundreds of μJ) and PRR (200 Hz to 16,000 Hz) of LEDs and also the advancements in low-noise data acquisition electronics. All these developments have resulted in commercialization of the technology, and it is worth mentioning that LED-PAI is now capable of functional imaging (oxygenation and blood flow imaging) of superficial and sub-surface tissue (more than 1 cm) at frame rates unachievable for laser-based systems (500 Hz). Even though LED-based PAI cannot be used for applications requiring larger imaging depth (for example, a full breast), it holds potential in several superficial imaging applications, specifically in rheumatology and dermatology.
Compared to solid-state lasers, the energy level of LED is two orders lower. However, the PRR of LED is much higher than that of a laser, which then can largely benefit the imaging quality by averaging more frames. Besides the emission energy level, another major difference between the pulsed LEDs and the solid-state lasers is the temporal pulse width. The pulse width of LED is tens of nanoseconds, whereas that of the solid-state lasers could be less than ten nanoseconds. The temporal pulse width imposes a limit on the spatial resolution of the imaging system. For example, the 70-ns pulse width of the 850-nm LED corresponds to a spatial resolution of 105 µm (=70 ns × 1500 µm/µs). This point, however, turns out to benefit the detection efficiency, especially when using bandlimited US probes for detection. The PA signal generated by a solid-state laser with a pulse width of 3.5 ns has a frequency component up to almost 300 MHz, in which anything above ~12 MHz will not be detected using a conventional mid-frequency range US probe. For a typical 5-MHz commercial US probe with 80% detection bandwidth, when using a 3.5-ns laser pulse, PA signal detection efficiency is 40 times less than the attainable efficiency when using an LED array generating 100-ns light pulses [
102]. The resolution offered by a typical 5–10-MHz clinical US probe is 200 µm. Therefore, the pulse width of the LEDs can potentially be extended to 100 ns without affecting the spatial resolution (70 ns setting is used in all the studies reported in this paper).
In the second part of this paper, we reviewed some of the preclinical and clinical applications reported using LED-based PAI (in sequence of phantom, ex-vivo, small animal, and human in-vivo studies) as listed in
Table 3. It is clear from the results that multispectral LED-based PA and US imaging holds strong potential in multiple applications, for example, guiding minimally invasive procedures, blood oxygen saturation imaging, diagnosis and staging of inflammatory arthritis, peripheral vascular assessment, guidance of surgical procedures like lymphaticovenous anastomosis, etc. In all the studies, an imaging depth of 0.5–1 cm was achieved at 10-Hz US and PA frame rates, which is good enough for multiple clinical applications. However, it is of paramount importance to improve the imaging depth for exploring more clinical applications, especially in the area of breast imaging and cardio-vascular medicine. To solve this issue and accelerate the clinical translation of LED-based PAI, several reconstruction and image processing techniques have been reported recently [
103,
104,
105]. Use of clinically approved contrast agents (for example, ICG) may also be also useful to enhance the imaging depth.
The LED-based PA imaging system described here (AcousticX) is safe for both skin and eye exposure. Since LED emissions are incoherent, the ANSI safety limits for collimated laser beams do not apply. Instead, the international electrotechnical commission (IEC) 62471 is followed. According to IEC 62471, the exposure limit for skin is based on thermal injury due to the temperature rise in tissue. Assuming that the illumination on the same skin area lasts continuously for 5 s using two 850-nm LED bars working at a 4-KHz pulse repetition rate, the estimated exposure is 4.57 × 103 W·m−2, which is below the thermal hazard limit for skin of 5.98 × 103 W·m−2. For eye safety, two aspects need to be considered, which are retinal thermal hazard exposure limit (weak visual stimulus) and infrared radiation eye safety limit. Assuming a continuous illumination at the front of the eye for 5 s using two 850-nm LED bars working at a 4-KHz pulse repetition rate, the estimated exposures are 2.92 × 103 W⋅m−2⋅sr−1 for retinal thermal exposure and 4.57 × 103 W⋅m−2 for infrared radiation exposure, both lower than the safety limits for eye of 1.34 × 105 W⋅m−2⋅sr−1 and 5.38 × 103 W⋅m−2, respectively.
To date, the maximum imaging depth achieved by LED-based PA imaging in an in-vivo setting is 1 cm. Even though this is encouraging considering the low pulse energy, it is important to improve LED optical output for better usefulness in more clinical applications. To an extent, signal averaging helps to improve SNR without effecting frame rate. However, if the magnitudes of the acoustic signals generated by the weak illumination are significantly below the noise-equivalent pressure level of the US probe, averaging will not effectively improve SNR. In terms of translational potential, addition of PA imaging to a clinical US system will have relatively easier clinical acceptance. Considering the acoustic bandwidth limitation of conventional pulse-echo US probes, we believe that the lower pulse width (70 ns is the setting used in all the applications reported in this paper) of LEDs is not a bottleneck. We foresee work on two aspects to improve the imaging depth: (1) use of nanostack technology to squeeze more light out of LED elements, with multiple p-n junctions embedded in the epitaxial layer, increasing the region’s light generation and thereby leading to a higher optical output, and (2) improving the driving electronics to increase the PRR further, consequently resulting in the possibility to average more frames and improve SNR without losing temporal resolution. LED might compete with LD with market share during clinical translation, but will not have much overlap with Nd: YAG OPO-based PAI system which focus on deep-tissue applications. However, LED-PAI may be a suitable option for point-of-care applications like guidance of peripheral vascular access procedures, rheumatoid arthritis screening, PWS diagnosis and treatment monitoring, etc.
Looking into the future, it is foreseen that advances in high-power LED technology, mainly driven by the lighting industry, and significant developments in machine/deep learning and signal processing algorithms will increase the use of LEDs in the context of PAI. LED arrays in different shapes could be also developed to find different applications, for example, developing ring-shape LEDs for breast 3D imaging and coupling light into optical fibers for minimally invasive and endoscopic procedures. Worth mentioning here, if sufficient focusing can be achieved, LEDs will be the ideal candidate for acoustic-resolution PA microscopy (PAM). There have also been some recent reports in this direction. In 2017, Dai et al. presented a PAM system based on miniature LEDs working at a 405-nm source, which showed the capability of in-vivo mapping of vasculature networks in biological tissue [
106,
107]. They used a high-power LED (power~1.2 W) working at 405 nm wavelength, and a pulse width of 200 ns. The repetition rate was extended to 40 KHz towards meeting the high demand of scanning speed in a PAM setting. They acquired a complete PAM image in vivo in about 1 h, which is an encouraging result for the first LED-based PAM proof-of-concept study, but this temporal resolution is not good enough for clinical studies. On the other hand, for optical resolution PAM, the use of LEDs is likely to be challenging, as it would be difficult to achieve the necessary micron-scale diffraction-limited spot sizes. It would be also interesting to develop new low-frequency US probes (2–3 MHz) with ultra-high bandwidth and sensitivity to achieve higher imaging depth without compromising spatial resolution.
5. Conclusions
The use of LEDs as an illumination source introduces some limitations. First, the LEDs cannot be spectrally tuned, which eliminates the possibility of PA spectroscopic applications in which multiple chromophores are involved. Second, the pulse width of the LED is low when compared to a laser (30–100 ns), which affects the stress confinement satisfaction and can impact the efficiency of acoustic wave generation. Third, LEDs have low optical output power—this can limit penetration depth at higher frame rates. However, LED-based PAI systems offers several advantages, including a significant reduction in cost, smaller footprint, no requirement of laser calibration and monitoring, and no need for optical goggles or light-tight shields. Thus, LED-based systems are not only suitable for point-of-care non-invasive applications complementing US imaging, but also ideal for personalized or wearable PA equipment, and we foresee that this technology could additionally have broad utility in a number of therapeutic drug monitoring applications.
With wide optical wavelength range, flexible pulse-width setting, small footprint, low cost, and energy efficiency, LED-based PAI holds strong potential in functional and molecular preclinical and clinical imaging. We foresee that the addition of LED-based PAI to conventional US imaging in a clinical scanner will have a huge impact in point-of-care diagnostic imaging and also accelerate the clinical translation of PAI.