Shortwave infrared diffuse optical wearable probe for quantification of water and lipid content in emulsion phantoms using deep learning

Abstract

Significance: The shortwave infrared (SWIR, $\sim 900$ to 2000 nm) holds promise for label-free measurements of water and lipid content in thick tissue, owed to the chromophore-specific absorption features and low scattering in this range. In vivo water and lipid estimations have potential applications including the monitoring of hydration, volume status, edema, body composition, weight loss, and cancer. To the best of our knowledge, there are currently no point-of-care or wearable devices available that exploit the SWIR wavelength range, limiting clinical and at-home translation of this technology.

Aim: To design and fabricate a diffuse optical wearable SWIR probe for water and lipid quantification in tissue.

Approach: Simulations were first performed to confirm the theoretical advantage of SWIR wavelengths over near infrared (NIR). The probe was then fabricated, consisting of light emitting diodes at three wavelengths (980, 1200, 1300 nm) and four source-detector (S-D) separations (7, 10, 13, 16 mm). In vitro validation was then performed on emulsion phantoms containing varying concentrations of water, lipid, and deuterium oxide ($D_{2}O$). A deep neural network was developed as the inverse model for quantity estimation.

Results: Simulations indicated that SWIR wavelengths could reduce theoretical water and lipid extraction errors from $\sim 6\%$ to $\sim 1\%$ when compared to NIR wavelengths. The SWIR probe had good signal-to-noise ratio ($>32\mathrm{dB}$ up to 10 mm S-D) and low drift ($<1.1\%$ up to 10 mm S-D). Quantification error in emulsion phantoms was $2.1\pm 1.1\%$ for water and $-1.2\pm 1.5\%$ for lipid. Water estimation during a $D_{2}O$ dilution experiment had an error of $3.1\pm 3.7\%$.

Conclusions: This diffuse optical SWIR probe was able to quantify water and lipid contents in vitro with good accuracy, opening the door to human investigations.

Keywords: deep learning; diffuse optical imaging; diffuse optics; lipid; shortwave infrared; water; wearable.