Regionally Discrete Aqueous Humor Outflow Quantification Using Fluorescein Canalograms

Purpose To visualize and quantify conventional outflow directly in its anatomic location. Methods We obtained fluorescein canalograms in six porcine whole eyes and six porcine anterior segment cultures. Eyes were perfused with a constant pressure of 15 mmHg using media containing 0.017 mg/ml fluorescein. Flow patterns were visualized using a stereo dissecting microscope equipped for fluorescent imaging. Images were captured every 30 seconds for 20 minutes for time lapse analysis. Anterior chamber cultures were imaged again on day three of culture. Canalograms were first analyzed for filling time per quadrant. We then wrote a program to automatically compute focal flow fits for each macropixel and to detect convergent perilimbal flow patterns with macropixels grouped into 3 equal-radial width rings around the cornea. A generalized additive model was used to determine fluorescence changes of individual macropixels. Results The resulting imaging algorithm deployed 1024 macropixels that were fit to determine maximum intensity and time to fill. These individual fits highlighted the focal flow function. In whole eyes, significantly faster flow was seen in the inferonasal (IN) and superonasal (SN) quadrants compared to the superotemporal (ST) and inferotemporal (IT) ones (p<0.05). In anterior chamber cultures, reduced flow on day 1 increased in all quadrants on day 3 except in IT (p<0.05). Perilimbal ring analysis uncovered convergent perilimbal flow. Conclusions An algorithm was developed that analyzes regional and circumferential outflow patterns. This algorithm found flow patterns that changed over time and differ in whole eyes and anterior segment cultures.


Introduction
This R Markdown document illustrates how to use the eye-canalogram package to quantitatively analyze canalograms obtained in aqueous-outflow eye models. R code is provided along with its output.

Load R Packages
The fist step is to load the required R packages and others used by this document:

Loading Images
The read.images function of the EyeCanalogram package is first used to read all canalogram images. The most important function parameter is the root parameter which is the common filename prefix for all images. For example, if root is ACs/d1/__13_/_13_T00, then all TIFF files beginning with _13_T00 in the ACs/d1/__13_ directory would be loaded. The n parameter can be used to restrict loading to the first n images found. The low parameter sets the number of macropixels used in each direction. Here low is 32 which will result in resizing all images to have 32 pixels in the smallest dimension (here the image height) with possibly more in the other dimension depending on the input image aspect ratio. cornea in all other canalogram images. If a file roi.tif exists, it too is converted to a black-and-white mask where white pixels indicate the perilimbal area and cornea to analyze -black pixels can be used to mask out regions not of interest such as the compression ring used for anterior chamber cultures.
For illustration, next the macropixels are extracted for a number of frames and displayed using ggplot2. The melt function from the reshape2 package is used to convert the 43 × 32 × 41 array of pixel information into a data.frame. The column names are then assigned, individual frames chosen, and finally those frames are displayed.

Raw Intensity
Macropixel Data

Process Images
The entire dataset is fit individually at each macropixel and all at once (global fit) with smoothing along the clock hours, radial distance, time, and a tensor product for the interactions. The fit.gam function of the EyeCanalogram package does all of the work. The function parameter r_bands sets how many radial bands around the cornea are used for the global fit and ring plots.

Quadrant Flow
Flow below is estimated from the increase in fluorescence (from the initial, possibly non-zero, fluorescence in the first frame) to the time at half-max. [This differs from the Filling Rate which is the is the increase in absolute fluorescence (from an initial value of zero) to the time at half-max.] The flow is determined for a thin region a given distance from the cornea. Here, a distance of 2 pixels from the cornea is chosen. The global-fit is used to estimate the fluorescence in this region and then flow is calculated. The proportion of the total flow for the entire analysis region (outside of the cornea and within the mounting ring) is then attributed to each quadrant. This can be converted into an estimate in microliters/minute given an estimated 3 microliters of total flow flow per minute for the entire eye.

Displaying Results
All of the data can be displayed at once as shown above. Alternatively, different measures of the fit can be displayed schematically.