Journal: Science Advances

Article Title: Machine reading and recovery of colors for hemoglobin-related bioassays and bioimaging

doi: 10.1126/sciadv.adt4831

Figure Lengend Snippet: ( A ) Detrimental color variations in digital photos of biological tissue captured under various white-light illumination conditions: light-emitting diodes (LEDs) with color temperature of 3000, 4300, and 5800 K, as well as fluorescent tube light. The colors under CIE illuminant E (equal energy radiator or spectrally uniform illumination) can be considered absolute. CIE illuminant E is achieved through spectral normalization using a diffuse (Lambertian) reflectance standard (see Materials and Methods). ( B ) Light conditions having distinct spectral profiles: fluorescent tube, incandescent light, white LED, and sunlight (fig. S1). ( C ) Representative photos of whole blood–mimicking samples in cuvettes at different hemoglobin (Hgb) concentrations, acquired under various light conditions. A conventional color chart (Macbeth ColorChecker or X-Rite ColorChecker) is juxtaposed with the samples. ( D ) Smartphone model–dependent RGB spectral response functions (also known as spectral sensitivity): Apple iPhone 12 Pro, Apple iPhone SE, Samsung Galaxy S21, and Samsung Galaxy A52 (fig. S2). ( E ) Representative photos captured using various smartphone models. ( F ) File formats with different bit depths (color depths) in the R, G, and B color channels: JPEG (8-bit depth), RAW (10-bit depth), and MP4 (8-bit depth). ( G ) Representative photo acquisition scenarios based on combinations of light conditions (B), smartphone models (D), and file formats (F). When multiple photos of the same sample are captured under varying conditions, accurate and precise color recovery ensures that recovered color values converge to the ground truth.

Article Snippet: Using a white diffuse reflectance standard (SRT-99-050, Labsphere), the spectral intensity I ref (λ) reflected from the white reflectance standard under the identical imaging setting as the sample can be obtained I ref ( λ ) = L ( λ ) · C ( λ ) · D ( λ ) (2) Then, O (λ) is calculated by normalizing I m (λ) with respect to I ref (λ) O ( λ ) = I m ( λ ) I ref ( λ ) (3) CIE illuminant E, through spectral normalization, allows for the definition of the absolute colors of a sample, as the spectral intensity is not influenced by the physical illumination source or acquisition conditions.

Techniques:

Journal: Science Advances

Article Title: Machine reading and recovery of colors for hemoglobin-related bioassays and bioimaging

doi: 10.1126/sciadv.adt4831

Figure Lengend Snippet: ( A ) Macbeth ColorChecker containing 24 reference colors used for general photography. ( B ) Corresponding CIE xy chromaticity values under CIE illuminant E, measured using a spectrometer and a reflectance standard. The wide gamut of Macbeth ColorChecker overlaps with the sRGB color space. ( C ) Corresponding CIE LAB values under CIE illuminant E on the a * and b * axes. ( D ) Corresponding L* values as functions of a* and b* values. ( E and F ) Parametric spectral modeling of biological tissue (peripheral tissue and blood samples). Physiologically possible color variations are captured by 12,240 synthesized spectral data of peripheral tissue (E) and 10,000 synthesized spectral data of whole blood (F) (see Materials and Methods). ( G ) Blood Hgb gamut defined by three primary points of CIE xy chromaticity: ( x , y ) = (0.30, 0.31), (0.47, 0.42), and (0.63, 0.33). ( H ) Corresponding CIE LAB values on the a * and b * axes. ( I ) Corresponding L* values as functions of a* and b* values. ( J and K ) Importance of CIE XYZ Euclidean distance metric for machine readability and learning in color-based diagnostics, compared to Delta E values including CIE94 ( ∆ E 94 * ) and CIEDE2000 ( ∆ E 00 * ). Eleven representative colors are selected from the Hgb gamut, with equal CIE XYZ Euclidean distances between all pairs of adjacent colors. Delta E values incorporate human visual judgment and perception.

Article Snippet: Using a white diffuse reflectance standard (SRT-99-050, Labsphere), the spectral intensity I ref (λ) reflected from the white reflectance standard under the identical imaging setting as the sample can be obtained I ref ( λ ) = L ( λ ) · C ( λ ) · D ( λ ) (2) Then, O (λ) is calculated by normalizing I m (λ) with respect to I ref (λ) O ( λ ) = I m ( λ ) I ref ( λ ) (3) CIE illuminant E, through spectral normalization, allows for the definition of the absolute colors of a sample, as the spectral intensity is not influenced by the physical illumination source or acquisition conditions.

Techniques: Synthesized

Journal: Science Advances

Article Title: Machine reading and recovery of colors for hemoglobin-related bioassays and bioimaging

doi: 10.1126/sciadv.adt4831

Figure Lengend Snippet: ( A ) One-shot transduction learning of neural network–based color recovery with HemaChrome. The neural network is trained for each photo without relying on any preexisting training dataset. The training dataset consists of the color values of the reference colors in HemaChrome. Once trained on the specific photo, the network processes the RGB values acquired from the sample of interest in the photo to recover the corresponding CIE XYZ values. ( B ) HemaChrome chart with 116 reference colors for neural network–based color recovery. ( C ) Corresponding CIE xy chromaticity values under CIE illuminant E, measured using a spectrometer and a reflectance standard. ( D ) Corresponding CIE LAB values under CIE illuminant E on the a * and b * axes. ( E ) Corresponding L* values as functions of a* and b* values. ( F ) Representative photo of blood Hgb–mimicking samples to recover their absolute colors (under CIE illuminant E). ( G to J ) Average color differences between the ground truth and recovered CIE XYZ values for each test sample from photos captured across 36 diverse photo acquisition scenarios . The root mean square error (RMSE) (G), root mean square relative error (RMSRE) (H), average CIE94 ( ∆ E 94 * ) (I), and average CIEDE2000 ( ∆ E 00 * ) (J) are compared (eqs. S1, S2, S6, and S7). Among the three color correction methods, neural network color recovery using HemaChrome consistently returns minimal errors across all test samples.

Article Snippet: Using a white diffuse reflectance standard (SRT-99-050, Labsphere), the spectral intensity I ref (λ) reflected from the white reflectance standard under the identical imaging setting as the sample can be obtained I ref ( λ ) = L ( λ ) · C ( λ ) · D ( λ ) (2) Then, O (λ) is calculated by normalizing I m (λ) with respect to I ref (λ) O ( λ ) = I m ( λ ) I ref ( λ ) (3) CIE illuminant E, through spectral normalization, allows for the definition of the absolute colors of a sample, as the spectral intensity is not influenced by the physical illumination source or acquisition conditions.

Techniques: Transduction

Journal: Journal of Biomedical Optics

Article Title: Development and characterization of a combined fluorescence and spatial frequency domain imaging system for real-time dosimetry of photodynamic therapy

doi: 10.1117/1.JBO.30.S3.S34103

Figure Lengend Snippet: Schematic of the combined SFDI and fluorescence imaging system for aPDT dosimetry. (a) Projection arm of the system provides spatially modulated light at four different wavelengths (395, 545, 625, and 850 nm) using two independent amplitude masks. The light is spatially combined and passed through a removable linear polarizer before being projected onto the sample, indicated with an orange arrow, at an angle. (b) Separately, the treatment arm delivers the light to the sample for PDT treatment and fluorescence excitation. The resulting reflectance and fluorescence signals are collected by the imaging arm at three different channels, λ < 590 nm , 590 nm ≤ λ ≤ 660 nm , and λ > 660 , where the middle channel is further cleaned up with a bandpass filter. (c) Picture of the combined imaging system. (d) Pictures of the printed amplitude masks used for 0.3 mm − 1 (left) and 1.0 mm − 1 (right) patterned illumination. (e) Normalized intensity profiles of the four projection LEDs at their respective detectors. ACL, aspheric condenser lens; DM, dichroic mirror; LP, linear polarizer; AD, achromatic doublet; BPF, bandpass filter; HAT, Hastings achromatic triplet.

Article Snippet: The spectral throughput of the system was assessed by measuring the spectral profiles of each LED at their corresponding detectors [ ] using a miniature spectrometer (Flame-S-VIS-NIR, Ocean Insight, Orlando, Florida, United States), with the light from the illumination arm being reflected off of a 99% diffuse reflectance standard (USRS-99-020, LabSphere, North Sutton, New Hampshire, United States) to allow for collection by the imaging arm without further influencing the spectral response.

Techniques: Fluorescence, Imaging

Journal: Journal of Biomedical Optics

Article Title: Development and characterization of a combined fluorescence and spatial frequency domain imaging system for real-time dosimetry of photodynamic therapy

doi: 10.1117/1.JBO.30.S3.S34103

Figure Lengend Snippet: Predicted spatial frequency–dependent reflectance of the human skin. SFDI Monte Carlo simulations were performed using a seven-layer human skin model (a) with a fixed width (1 mm) and depth (3 mm) and a variable length such that two full periods of the patterned illumination were always applied. The thickness ( d , mm), scattering anisotropy ( g ), and refractive index ( n ) for each layer were fixed, whereas the absorption ( μ a , mm − 1 ) and reduced scattering ( μ s ′ , mm − 1 ) coefficients were varied for the four illumination wavelengths. The absolute (b) and normalized (c) Monte Carlo results for the reflectance of the seven-layer model with spatial frequencies between 0 and 1.0 mm − 1 at the four illumination wavelengths.

Article Snippet: The spectral throughput of the system was assessed by measuring the spectral profiles of each LED at their corresponding detectors [ ] using a miniature spectrometer (Flame-S-VIS-NIR, Ocean Insight, Orlando, Florida, United States), with the light from the illumination arm being reflected off of a 99% diffuse reflectance standard (USRS-99-020, LabSphere, North Sutton, New Hampshire, United States) to allow for collection by the imaging arm without further influencing the spectral response.

Techniques: Refractive Index

Journal: Translational Vision Science & Technology

Article Title: Artificial Intelligence–Driven Detection of LASIK Using Corneal Optical Coherence Tomography Maps

doi: 10.1167/tvst.14.3.17

Figure Lengend Snippet: Descriptive Statistics of Corneal Optical Coherence Tomography Measurements

Article Snippet: The corrected OCT signal was then converted to a reflectance scale calibrated by a standard diffuse reflectance target (Spectralon, Labsphere, North Sutton, NH, US).

Techniques: Tomography