Work Description

Authors:
Joshua M. Herzog
University of Michigan
Mechanical Engineering
jmherzog@umich.edu
0000-0001-9089-819X

Angela Verkade
University of Michigan
Ophthalmology and Visual Sciences
0000-0002-7033-4853

Volker Sick
University of Michgan
Mechanical Engineering
0000-0001-5756-9714

Created: July 31, 2024

Introduction:
Quantitative fluorescence imaging has been investigated for the development of
low-cast, rapid, and non-invasive imaging of anterior segment disease. In
particular, quantitative fluorescence imaging is being investigated to detect
corneal infections, to quantify corneal aberrations and disease, and to
quantify damage and chemical changes in the lens. Two of these aspects are
explored in the following articles:

Herzog, Joshua M., Verkade, Angela, and Sick, Volker. "Corneal shadowgraphy:
a simple, low-cost, rapid, and quantitative tool with potential clinical
utility." Manuscript submitted to Communications Medicine. 2024.

Herzog, Joshua M., Verkade, Angela, and Sick, Volker. "Quantitative and rapid
in vivo imaging of human lenticular fluorescence." Manuscript submitted to
Investigative Ophthalmology and Visual Science. 2024.

The data in this repository is the first set of quantitative fluorescence
images used in these two approaches.

Review Board Approval:
Institutional Review Boards of the University of Michigan Medical School (IRBMED)
Approval #HUM00210338
Approved on: November 30, 2023

Methodology:
Fluorescence of human eyes was investigated in a population of 30 generally
healthy adults. The study was approved (approval #HUM00210338) and overseen by
the Institutional Review Boards of the University of Michigan Medical School
(IRBMED). All activities were conducted exclusively on the University of
Michigan's College of Engineering Campus in Ann Arbor, Michigan, and conformed
to the ethical standards of IRBMED. Participants were recruited exclusively
online using the University of Michigan's UMHealthResearch.org webportal, which
is an IRBMED-approved electronic recruiting platform, and all participants
provided comprehensive, written informed consent prior to participation. The
study was open to adults age 18 to 65. Photosensitive individuals, individuals
taking photosensitive medications, and individuals with acute eye disease were
excluded from the study.

Fluorescence images of the 60 human eyes (30 participants, 2 eyes per
participant) were collected using a custom fluorescence imaging system. A near-
UV LED (365 nm, 2 W peak power) is collimated and directed onto the eye with a
peak irradiance of 30 mW/cm2 and for a duration of 50 ms. The UV source excites
fluorescence from chemicals inside the eye, especially in the lens. The
fluorescence emission is collected by an achromatic doublet lens and split
between two image sensors. The "Blue" band captures emission of approximately
400 to 450 nm, and the "Red" band captures emission above 450 nm which
typically extends to around 600 nm. A series of calibration images are
collected to provide a quantitative reference, including images of a
fluorescent glass window and background/black-reference images. A detailed
radiometric analysis of the imaging setup is provided in:

Herzog, J. M., & Sick, V. (2023). Design of a line-of-sight fluorescence-based
imaging diagnostic for classification of microbe species. Measurement Science
and Technology, 34(9), 095703.

Data Description:
Data deposited here includes 60 image sets (30 individual participants, and 2
eyes per individual) consisting of raw fluorescence images, diffuse reflection
images using ambient lighting, images used for correction, and calibration, and
metadata. Images are split into two wavelength bands as described in the
methodology. Image sets are stored in Hierarchical Data Format 5 (HDF5) files
with the following nodes:

Red-band fluorescence image: /Raw/Fluorescence/Red
Red-band fluorescence background image: /Raw/Fluorescence/RedRef
Blue-band fluorescence image: /Raw/Fluorescence/Blue
Blue-band fluorescence background image: /Raw/Fluorescence/BlueRef

Red-band reference image: /Calibration/Profile/Red
Red-band reference background image: /Calibration/Profile/RedRef
Blue-band reference image: /Calibration/Profile/Blue
Blue-band reference background image: /Calibration/Profile/BlueRef

Red-band diffuse reflection images: /Raw/Diffuse/Red
Blue-band diffuse reflection images: /Raw/Diffuse/Red

Additionally, each image node contains a tag for frame rate, exposure duration,
and timestamp.

The approved informed consent document associated with this study is included
in "InformedConsentForm.pdf". This form is exactly as approved by IRBMED and
as signed by all participants.

Summary statistics including demographic data, participant-reported diseases
(e.g., diabetes, keratoconus), and pupil size are also stored in a Microsoft
Excel spreadsheet file, and again as a text-file for easier access. Finally,
2-channel pseudocolor images combining the two fully-processed image bands are
stored as portable network graphics (PNG) files. Image processing here includes
background and flatfield correction, image registration, pupil detection, and
2x2 software binning. Image processing methods are described in detail in:

Herzog, J. M., & Sick, V. (2024). Fluorescence imaging for the anterior segment
of the eye. Frontiers in Photonics, 4, 1336541.

The corrected intensity at pixel i of image j for patient k is calculated as

I(i,j,k) = (I_raw(i,j,k) - ) / ( - ) * I_std

where the angle brackets <> denote an average and the subscripts "raw", "bg",
"ref", and "ref,bg" refer to the raw, background, reference, and reference
background images, respectively. In the above notation, the average is taken over
the missing subscript. The subscript "std" refers to a fluorescence standard,
which here is taken to be equal to

I_std = -

or rather, the reference image set for patient #1 is used. Note that the
standard is a number (rather than an array or image) as it is averaged
over multiple pixels as well. In this case, the average is taken over the
center 50x50 square.