The Caltech-Twente team found that, owing to wave-particle duality, even as few as 1,000 signal photons, scattered over 220,000 camera pixels, can be sufficient to image within scattering tissues using optical phase conjugation. Because of a surprising feature of interferometric detection, the authors suggest, each signal photon contributes some information to all individual camera pixels simultaneously, adding a small statistical bias toward the correct measurement. [Image: Courtesy of I.M. Vellekoop]
Doing microscopy deep in turbid, strongly scattering media, such as biological tissues, must always grapple with the issue of “photon budget.” How many photons must be retrieved from scattering tissue to correct for the impact of the scattering, and actually extract a useful signal amid the noise? The intuitive answer is at least one photon per camera pixel—and, ideally, a lot more.
For one particular imaging technique, however, digital optical phase conjugation (DOPC), scientists from the California Institute of Technology (Caltech), USA, and the University of Twente, Netherlands, have come up with a striking and counterintuitive alternative response. Their experiments suggest that, given a detector of sufficiently high resolution, a mere handful of photons—as few as 0.004 photons per detector pixel—can give sufficient information for DOPC imaging in scattering media (Phys. Rev. Lett., doi: 10.1103/PhysRevLett.118.093902). The researchers believe that the finding could substantially expand DOPC’s horizons in biomedical applications.
“Time reversing” scattering’s impact
DOPC works by “time reversing” the scattering of light. Central to the technique is establishing a specific, labeled point, called a “guide star,” at a known location in the sample. In some cases the guide star is an actual light source, such as a fluorescent particle, physically placed within the sample at a given spot. The more common approach, however, is to use a focused, external ultrasound beam that frequency-shifts light propagating through the guide-star point, and thus allows the guide-star point to behave as a frequency-modulated, virtual light source.
The DOPC process begins with a so-called recording step, in which the guide star is illuminated by a probe beam. The phase and amplitude of the frequency-modulated scattered light from the guide star are measured interferometrically with a second, reference beam. Then, in the playback step, a phase-conjugate copy of the light field is created (commonly with a spatial light modulator, or SLM) using that information. This field, sent back through the scattering medium, backs out the effect of the scattering, yielding sharp focus at the guide star’s known position.
The problem has been that, owing to the scattering, guide stars commonly provide very few photons during the recording step, and the minimum photon budget required to create a phase-conjugate version of the scattered light from the guide star hasn’t been well understood. That has held back DOPC as a mechanism for imaging deep within tissue, or for high-speed imaging, where photon budgets tend to be extremely low.
A mere thousand photons
OSA Fellow Changhuei Yang and colleague Mooseok Jang at Caltech, along with OSA Member Ivo Vellekoop at the University of Twente, decided to dig into the question of just how many guide star photons were needed to make DOPC work. They set up a DOPC experiment to image a thin (450-µm) layer of opal, known to be a highly scattering material.
Next, they subjected the sample to probe beams of varying intensities in the recording step, interfering the scattered light with a reference beam and capturing the interfered field distribution on a 1920×1080-pixel CMOS sensor. They then sent the playback beam through an SLM to create a phase-conjugated wavefront to re-illuminate the sample and captured the result on a CCD camera.
“When we started the project,” notes Vellekoop, “we naively expected that each pixel of the camera would need to detect at least a couple of scattered photons to have a useful signal.” What they found instead was that they could obtain a good recording even when there were far fewer photons than available pixels—as few as 0.004 photons per pixel, or around 1,000 photons for a 220,000-pixel detector. It appears, says Vellekoop, that when the photons are detected interferometrically, each photon can, ever so slightly, bias the measurement outcome at all 220,000 detector positions at once.
Such a decidedly non-quantized view of photons, the team’s analysis suggests, likely arises from wave-particle duality. Because quantization occurs after interferometric mixing of the wavefronts, the wavefront information encoded in the photons can be useful in “steering” the output, even though each pixel samples less than a full quantum of signal energy on average. It is, Vellekoop notes, “a somewhat counterintuitive finding that may force many physicists to rethink their image of what a photon actually is.”
The research team believes that the findings, in addition to shedding light on the fundamental issue of how interferometric techniques behave in the low-photon limit, could prove practically useful for applying DOPC to high-speed and deep-tissue imaging, for which a low photon budget from the guide star is often the limiting factor. And, according to Vellekoop, the finding could also open up other opportunities—for example, for focusing light deep within brain tissues to control neurons in optogenetic studies.