Seeing Inside the Body With Red Light

Shine a red laser pointer through your finger and it glows. That glow is the whole idea. Red and near-infrared light pass into the body far more easily than most people assume. It’s harmless, cheap to produce, and already hiding in devices you own. The problem is that it doesn’t travel in a straight line. It scatters, bouncing off cells millions of times until the image it carries turns into fog.

For a century, that fog made light a surface-only tool for looking inside people. That’s starting to change, and I think it’s one of the most underrated opportunities in healthcare.

The imaging gap

An MRI is a marvel: a superconducting magnet, liquid helium, a shielded room, stable power, trained technicians and radiologists to read the results. All of that makes it expensive, and it concentrates the machines where money and infrastructure already are.

The gap is enormous. Surveys of MRI availability put Africa at roughly 0.7 scanners per million people in 2017, the lowest of any WHO region, and some countries had none at all. West Africa had about 84 units serving more than 370 million people. Compare that with dozens per million in Japan and the United States.

Even in wealthy countries, an MRI is something you travel to and wait for. It isn’t something that comes to the patient at the roadside, the bedside or the village clinic.

Why light works at all

Biological tissue has an optical window. Between roughly 650 and 950 nanometers (deep red into near-infrared), water and blood absorb relatively little, and light can travel several centimeters into tissue. That’s why pulse oximeters work. It’s also the basis of established techniques like near-infrared spectroscopy, which tracks oxygenated and deoxygenated blood in the brain and muscles.

The catch is scattering. The light gets in, but the picture gets scrambled. Three ideas together are starting to unscramble it.

Idea 1: undo the scattering

Scattering looks random, but it’s deterministic: the same tissue scrambles the same light the same way. In 2007, Ivo Vellekoop and Allard Mosk showed that by shaping the incoming wavefront with a spatial light modulator, you can pre-compensate for the scrambling and focus light to a sharp point through a strongly scattering material. The field that grew out of that work, wavefront shaping, borrows heavily from holography: measure how the medium distorts light, then send in the inverse.

Living tissue moves, which makes this hard: blood flows and cells shift. But the principle holds. The fog is a code, not noise.

Idea 2: measure what light does, not just what it shows

You don’t always need a picture to save a life. Openwater has built a portable device that sends short pulses of near-infrared laser light into the head. It reads the resulting speckle pattern with camera sensors and measures blood flow in the brain during a 70-second bedside scan. In a published clinical study, it detected large-vessel-occlusion strokes with about 79% sensitivity and 84% specificity, outperforming the standard stroke scales emergency crews use today.

That matters because large-vessel strokes need a specific kind of hospital, and every minute of misrouting costs brain tissue. A device like this doesn’t replace a CT or MRI. It makes a faster, better triage decision before one is available.

Idea 3: let AI reconstruct the image

Turning scattered light back into an image, known as diffuse optical tomography, is a notoriously hard inverse problem: many internal structures could explain the same measurements. Classical solvers are slow and fragile. Deep learning is changing that. Recent research shows neural networks reconstructing optical images with better quality and dramatically faster, in one reported case from 3.8 minutes to 0.02 seconds per reconstruction. That’s the difference between a research instrument and something you can use at the bedside.

The fog is a code, not noise, and AI is getting very good at reading codes.

What’s proven, and what isn’t yet

Being clear about this matters, because health technology has been burned by overpromising before.

• Proven: near-infrared light safely reaches centimeters into tissue. Optical methods measure blood oxygen and blood flow in the clinic. Portable optical stroke triage has promising clinical data. Wavefront shaping works in the lab. Deep learning speeds up optical reconstruction.
• Not yet: MRI-level resolution deep inside the body. It may never happen for deep organs, and that’s fine. Light’s resolution degrades with depth, so optical imaging is most promising in the first few centimeters: the brain’s surface through the skull, breast tissue, infants, limbs, wounds and perfusion.

The honest pitch isn’t “light replaces MRI.” It’s that light answers the questions that don’t need an MRI, in the places that don’t have one.

What I’d build

If I were leading this product, I’d make four decisions early.

1. Start with one life-or-death triage question, not a general imager. Stroke is the model: one clinical decision (which hospital?), a clear time window, and an outcome anyone can measure. Other candidates include brain bleeds after head trauma, blood flow in diabetic feet, and breast screening where mammography isn’t available.
2. Design for the clinic with no radiologist. The output shouldn’t be an image for an expert to interpret. It should be a decision support signal with a confidence level, built for a nurse or paramedic, with the raw data sent to a remote specialist when connectivity allows.
3. Treat cost as a feature. Laser diodes and camera sensors are consumer-electronics parts. The target should be a device that costs about as much as an ultrasound machine, not an MRI. It should run on a battery and hold up in heat and dust.
4. Plan the regulatory path from day one. In the US, most devices like this go through the FDA’s 510(k) or De Novo pathways. A narrow triage claim is a faster, more credible path than a broad imaging claim, and each approved use funds the next.

Medicine’s most powerful tools tend to be its most concentrated. Red light points the other way: harmless, cheap and portable, with AI supplying what the hardware can’t. I’d like to help build it.

---