Draft:Detection of Optically Luminescent Probes using Hyperspectral and diffuse Imaging in Near-infrared

• Comment: This is an advertising page on a 2019 paper which has 99 cites. Note that the claim that it was published "in Nature" is incorrect and misleading, Wikipedia is not a place to advertise your work. Ldm1954 (talk) 08:33, 24 April 2025 (UTC)

• Comment: May be notable(but needs refs), Ozzie Ozzie10aaaa (talk) 13:28, 15 April 2025 (UTC)

---

The Detection of Optically Luminescent Probes using Hyperspectral and diffuse Imaging in Near-infrared (DOLPHIN) is an optical imaging system designed to noninvasively detect fluorescent probes in biological tissues. The system combines two key technologies: hyperspectral imaging (HSI) and hyperdiffuse light analysis (HDI) in the second near-infrared (NIR-II, 1000–1700nm) window to achieve high imaging depths while maintaining high spatial resolution. DOLPHIN was developed by Xiangnan Dang and a team of researchers at MIT, their work published in Nature in 2019.

System Overview

DOLPHIN employs a trans-illumination configuration, where excitation light is delivered from below the specimen and fluorescence is detected from above using a liquid nitrogen-cooled InGaAs detector. The system alternates between two modes:

• Hyperspectral Imaging (HSI): Captures spatially resolved fluorescence spectra across 800–1700 nm.
• Hyperdiffuse Imaging (HDI): Records spatial diffusion profiles of emitted light using bandpass filters corresponding to key spectral regions identified during HSI.

Together, these modes enable the extraction of both spectral and spatial features necessary for reconstructing fluorescence distributions at depths previously unattainable with conventional systems.

System Architecture

Design schematic of HSI (top) and HDI (bottom) configurations

DOLPHIN is comprised of a several key components:

• Excitation source: DOLPHIN utilizes a collimated laser to illuminate its target, which first passes through lenses and mirrors to adequately size and direct the beam.
• Hyperspectral Camera: A liquid nitrogen-cooled InGaAs camera is used to capture IR spectra.
• Monochromator (HSI mode only): A pair of convex lenses are used to collect light from the surface of the sample to the entrance of a slit of a monochromator with NIR diffraction gratings to split the NIR signal to resolve wavelengths. This is removed in HDI mode so that the distribution of diffuse light at selected wavelengths can be measured.
• Optical Filters: While both setups incorporate filters between the sample and the camera to prevent direct or scattered light from interfering with the signal from the sample, the HDI setup features additional bandpass filters to select for specific spectral band components.
• Silicon Camera: Takes bright field images that are useful for final 3D reconstructions

Mathematical Background and Image Processing Pipeline

Initial Processing

Both HSI and HDI mode start by obtaining a raw image that captures data in a 200x100 spatial grid (x,y) that represents the position of the object and a 320x256 intensity grid (a, b) that represents the position of the camera's detectors. This image is transformed to a 320 band hyperspectral cube image, I (x, y, λ), where λ is wavelength.$$I(x,y,a,b)\rightarrow I(x,y,\lambda ,b)\rightarrow I(x,y,\lambda ):HSC(x,y,\lambda )$$Next, a principal component analysis (PCA) algorithm is used to decompose the hyperspectral data into spectral contributions (Coeff), spatial contrast images (Score), and variance contributions (Explained), allowing identification of key spectral bands and separation of probe signals from tissue autofluorescence using the following formula.

$\left[\mathrm {Coeff} (\lambda ),\mathrm {Score} (x,y),\mathrm {Explained} \right]=\mathrm {PCA} \left(HSC(x,y,\lambda )\right)$

This process divides signals into four spectral bands. The α to δ bands are defined spectral regions separating excitation light (α), high wavenumber tissue autofluorescence (β), low wavenumber tissue autofluorescence (γ), and clean probe emissions (δ) to help distinguish signals and improve imaging contrast.

Flowchart of bioimaging procedures with DOLPHIN

HSI Specific Processing

For HSI, intra-band analysis is achieved for each pixel, providing information on spectral intensity (SI), spectral position (SP), and spectral width (SW). SI represents the total number of photon counts at each position for each spectral band and conveys contrast information. SP is the center emission wavelength at a point and is computed as the point at 50% of the accumulated spectral area, which is useful in depth estimation. SW is the broadening of the emission spectrum at each point, and is computed by subtracting the wavelength at 85% spectral accumulation from the wavelength at 15% accumulation. Broader SW values are correlated with more scattering, so SW can often help distinguish probe signals from the background. These quantities are mathematically represented via the following equations:

$\mathrm {SI} _{i=\alpha -\delta }(x,y)=\sum _{\lambda }\mathrm {HSC} _{i=\alpha -\delta }\left(x,y,\lambda (i)\right)$

$\mathrm {SP} _{i=\alpha -\delta }(x,y)=\mathrm {Spectral\ Position} \left(\mathrm {HSC} _{i=\alpha -\delta }\left(x,y,\lambda (i)\right)\right)$

$\mathrm {SW} _{i=\alpha -\delta }(x,y)=\mathrm {Spectral\ Width} \left(\mathrm {HSC} _{i=\alpha -\delta }\left(x,y,\lambda (i)\right)\right)$

In a final step, the SI ratio at different spectral bands is computed, which provides further information on image contrast and object visualization that allows for 3D reconstruction with the following equation:

$\mathrm {SI} _{i/j}={\frac {\mathrm {SI} _{i=\alpha -\delta }(x,y)}{\mathrm {SI} _{j=\alpha -\delta ,\,j\neq i}(x,y)}}$

HDI Specific Processing

For HDI, after spectral bands are defined, the hyperdiffuse data cube, HDC(x, y, r), is generated by converting detector pixel positions a and b into a radial distance r. Next, a different PCA method generates information on scattering patterns.

$\left[\mathrm {Coeff} ,\ \mathrm {Score} ,\ \mathrm {Explained} ,\ \mu \right]=\mathrm {PCA} \left(\mathrm {HDC} _{\alpha -\delta }(x,y,r)\right)$

Coeff defines how principal components are derived from diffuse profiles, Score provides the contrast images from their linear combinations, and Explained quantifies each component's contribution to the overall hyperdiffuse data cube. This PCA is applied to perform extraction of Diffuse Intensity (DI) and Scattering Radius (SR)—the latter serving as another measure of depth, since photons from deeper sources scatter more before reaching the detector. DI and SR are then used to perform the 3D reconstruction.. By fusing bright field images with information from both modes—HSI's spectral depth cues and HDI's spatial diffusion patterns—DOLPHIN reconstructs high-resolution 3D fluorescence maps using only a single 2D detector array as opposed to a 360 detector array. Notably, this reconstruction does not require any prior knowledge of the tissue's optical properties or the probe's spectral signature

Calculation of Theoretical Penetration Depth

Modeling penetration depth of DOLPHIN can be achieved if there is an estimate of optical properties of the object and surrounding environment. In HSI mode, depth of the fluorescence signal at a specific position is derived using Beer's Law:

$\left[\mathrm {Coeff} (\lambda ),\mathrm {Score} (x,y),\mathrm {Explained} \right]=\mathrm {PCA} \left(HSC(x,y,\lambda )\right)$

Where I₀ and I, are the intrinsic fluorescence intensity of the probe at zero depth of penetration and measured fluorescence intensity through tissue, respectively, which are linearly related through and µ_eff, the effective attenuation coefficient of tissue. Signal depth, d, is obtained through this equation. In HDI mode, the scattering profile is assumed to have cylindrical symmetry, I(r), for uniform objects. It is also assumed that the light source emits a spherical wave in a homogenous optical medium, which leads to the equation:

$I(r)\cdot \left(r^{2}+d^{2}\right)\cdot e^{\mu _{\mathrm {eff} }{\sqrt {r^{2}+d^{2}}}}=\mathrm {constant}$

This is used to fit d and µ_eff. For nonsymmetric scattering cases, the equation becomes:

$I(a,b)\cdot \left(r^{2}+h^{2}\right)\cdot e^{\mu _{\mathrm {eff} }{\sqrt {r^{2}+h^{2}}}}=\mathrm {constant}$

where $r^{2}=\left(a-a_{0}\right)^{2}+\left(b-b_{0}\right)^{2}$ and $\quad h=d-z(a_{0},b_{0})+z(a,b)$; a and b are the in-plane spatial coordinates; z is the height at each pixel; and a₀ and b₀ are the center coordinates of the incident beam. Both d and µ_eff can be fitted from the fluorescence signal and height profile, allowing for reconstruction.

Performance and Applications

DOLPHIN demonstrated the ability to detect fluorescent clusters as small as 100 µm through up to 8 cm of tissue phantom and 6 cm of muscle tissue. In vivo, it tracked the transit of near-cellular-sized probes through the gastrointestinal tract of living mice. Furthermore, the system's label-free background suppression and compatibility with FDA-approved contrast agents make it particularly promising for clinical translation compared to radiation-based modalities or methods requiring specialized tracers. DOLPHIN also holds promise in several biomedical applications including:

Early Cancer detection:

DOLPHIN enables noninvasive detection of tumors as small as 0.1 mm (≈200 cells) through 8 cm of tissue-mimicking phantom and 6 cm of muscle tissue, surpassing the resolution and depth limits of conventional modalities like MRI (≈500 μm resolution) and CT (≈600 μm resolution).

Drug delivery monitoring:

The system can track 100 μm-sized fluorescent probes in live animals, enabling:

1. Pharmacokinetic Studies: Noninvasive observation of drug distribution and clearance.
2. Targeted therapy evaluation: Assessing nanoparticle delivery efficiency to tumors or immune cells.

Gastrointestinal and Immune Cell Tracking:

DOLPHIN demonstrated real-time monitoring of a 100 μm probe through the entire GI tract of a living mouse, a feat unachievable with existing optical systems. This capability is critical for studying:

1. Inflammatory Bowel Disease: Tracking immune cell migration.
2. Microbiome Interactions: Observing bacterial or nanoparticle transit.