Optical and Spectroscopic Glucose Sensing

Optical and Spectroscopic Glucose Sensing

Optical glucose sensing estimates glucose from how it interacts with light, either by absorbing or scattering it or by changing how brightly a glucose-sensitive dye glows. Light could, in principle, read glucose through the skin with no needle. The record so far is sobering: a 2022 review of bloodless glucose monitors identified 28 noninvasive optical products and found that none had FDA clearance [1]. As of October 2026, US patients can get an optical CGM that works from under the skin: the Eversense 365, an implanted sensor that reads glucose through fluorescence [2].

What spectroscopy measures

Spectroscopy means shining light on a sample and recording which wavelengths come back changed. Molecules vibrate at characteristic frequencies, so glucose leaves its own pattern in a spectrum. Near-infrared (NIR), mid-infrared (MIR) and Raman spectroscopy, the three methods researchers return to most often, are all forms of this vibrational spectroscopy [3].

The hard part is scale. Glucose sits in blood and interstitial fluid (the fluid around cells just under the skin) at millimolar concentrations, well below the main components of skin, and the optical signal it produces is weak [1]. A device also has to tell glucose apart from proteins such as albumin and other small molecules in the same fluid [1]. Then the skin itself gets in the way: the review lists pigmentation, skin thickness, blood flow, body movement and room temperature among the things a noninvasive optical monitor has to account for [1].

Near-infrared: a faint signal in a sea of water

Short-wave NIR light can pass through thin tissue such as an earlobe, but in this range glucose shows up only through overtone and combination bands, weaker echoes of its main vibrations [3]. A 2021 review by Heise and colleagues describes a tiny glucose absorbance sitting on a large, shifting background from water, made worse by instrument drift [3]. So NIR systems depend on multivariate calibration, statistical models that pull the glucose pattern out of the whole spectrum [3].

Here's what surprised us. The same review argues that the deepest problem may not be the optics. Glucose levels in blood, in interstitial fluid and inside cells don't move in perfect step, so light that samples a mix of those compartments is chasing a moving target [3].

Mid-infrared: a clearer fingerprint, a shallow view

Between about 8 and 10 micrometers in wavelength, glucose absorbs light in a highly specific way, sometimes called its "glucose fingerprint" [4]. The cost is depth. Light in that range reaches only about 60 to 100 micrometers into skin, depending on how much water the outer layer holds [4].

DiaMonTech, a Berlin company, built a prototype that excites the skin with a mid-infrared quantum cascade laser and picks up the tiny amount of heat glucose releases, a method called photothermal detection [4]. In its study of 100 volunteers, measured at the finger over two hours, readings differed from a clinical glucometer by 12.1% on average [4]. Two caveats travel with that number. Company founders and employees ran the study, and a handheld version was still in development [4].

Raman: specific but quiet

Raman spectroscopy aims a laser at tissue and analyzes the small share of light that scatters back at shifted wavelengths. That scattering is weak, so the signal is poor. In return it is chemically specific and little troubled by water in tissue [5].

RSP Systems in Denmark tested a table-top Raman device that focuses at a "critical depth" below the skin's tough outer layer, in the interstitial fluid at the base of the thumb [5]. Thirty-five people with diabetes used it at home for 60 days, and its calibration stayed stable for more than 10 days [5]. Like the MIR study, this one was funded by the company developing the device [5].

Smartwatches and smart rings

In February 2024, the FDA warned against smartwatches and smart rings that claim to measure blood glucose without piercing the skin. The agency said it had not authorized any such device to measure or estimate glucose on its own [6]. Watch apps that display readings from an FDA-authorized CGM are a different product [6].

How the Eversense 365 reads glucose with light

The Eversense design avoids looking through skin by going under it. A healthcare provider inserts a sensor about 3.5 by 18.3 millimeters beneath the skin [2]. A fluorescent indicator grafted onto its surface binds glucose selectively and reversibly; when glucose binds, the indicator fluoresces more brightly, and the sensor's own optical system measures the change [2]. The sensor draws power by radio frequency from a transmitter worn on the skin above it, which turns the signal into a glucose value and sends a reading every 5 minutes to a phone app [2]. Our article on implantable fluorescence sensors (Eversense) covers the device in more detail.

There is no enzyme in this chemistry. Senseonics researchers described the indicator in 2014 as a fluorescent, boronic-acid based glucose-indicating polymer coated onto a miniature optical detection system, and the paper's title calls the system "abiotic" [7].

Long wear brought its own chemistry problem. A 2013 paper from Sensors for Medicine and Science reported that in human clinical testing of an implanted, glucose-sensitive hydrogel sensor, the boronate recognition element was oxidized off the fluorescent indicator, and the signal faded within hours [8]. The authors traced the damage to reactive oxygen species, mainly hydrogen peroxide, from the body's inflammatory response to the implant [8]. A 3-nanometer platinum coating on the hydrogel protected it in their tests and stretched the expected sensor life from 1 day to 6 months or more [8]. That is published research, not labeling; the FDA summary for the Eversense 365 doesn't mention platinum. It does describe a silicone collar that releases 1.75 mg of dexamethasone acetate, an anti-inflammatory steroid, to limit the local foreign body reaction [2].

The first Eversense, approved in June 2018 for up to 90 days [9], has since been replaced: Senseonics moved the implant to 180-day and then 365-day wear [10]. The Eversense 365 is cleared for up to 1 year in adults 18 and older, as an integrated CGM that can send readings to connected devices, including automated insulin delivery systems [2].

Accuracy and trade-offs of the optical implant

In the FDA review, 92.0% of Eversense 365 readings between 40 and 400 mg/dL fell within 20% of laboratory reference values [2]. The published ENHANCE study of the same system implanted the sensor in 110 adults and reported an overall mean absolute relative difference (MARD, the average percentage gap between sensor and reference readings) of 8.8%, mostly with one calibration a week; 90% of sensors lasted the full 365 days [10].

The FDA decision summary also lists the limits. Tetracycline-class antibiotics can falsely lower readings, while mannitol or sorbitol given intravenously, or in irrigation or peritoneal dialysis solutions, can falsely raise them [2]. The transmitter is MR Unsafe and has to come off before an MRI [2]. Calibration follows a set schedule: after a 24-hour warm-up, four fingerstick calibrations within 36 hours, then one a day through day 13, then one a week [2]. Whether an implanted sensor fits a person's diabetes care is a decision to make with their care team.

How electrochemical CGMs differ

The Dexcom G7 doesn't use light. It measures glucose with the enzyme glucose oxidase, which oxidizes glucose and produces hydrogen peroxide; the peroxide generates an electrical current that the system converts into a reading [11]. The FreeStyle Libre 3 Plus and Libre 2 Plus sensors also use glucose oxidase and measure a current proportional to glucose [12].

For light-based monitors that stay outside the body, the 2022 review names the central unsolved problem plainly: keeping measurements accurate over time is the major analytical challenge for every noninvasive optical technology [1]. Our article on non-invasive optical glucose sensing physics goes deeper into why.