Abstract
Purpose :
A custom computer system was used to generate a large amount of data to estimate evaporation, osmosis, and different types of flow within regions of TBU in tear films (TF) for normal subjects. A convolutional neural network (CNN) automatically identified TBU and non-TBU in fluorescent (FL) images gathered in vivo. FL intensity data in those regions was extracted for fitting by three ordinary differential equations (ODE) models for TBU. Model parameters were optimized to best fit the FL data in order to identify the mechanisms active within TBU regions.
Methods :
The FL intensity data was originally recorded from 25 normal subjects with 20 trials taken over two visits (Awisi-Gyau, Indiana University PhD thesis, 2020). We extract the experimental FL image data from the centers of TBU regions identified by the CNN. The data were fit with the ODE models using parameters representing evaporation rate (v), steady tangential flow (strain) rate (a), or decaying flow (b1) with decay rate (b2). A least squares minimization of the difference between experimental and computed intensities determined the parameters. Initial FL concentration and localized film thickness was estimated as in previous work (Wu et al IOVS 2015, 56:4211; Luke et al Bull Math Biol 2020, 82:71). All programs were custom using Python, Julia and/or Matlab.
Results :
Extraction resulted in N=467 usable instances of TBU from 15 subjects. Evaporation rates fall near or within experimental ranges. Statistical distributions of thickness and osmolarity were computed for individual subjects and for the population. Findings include: (i) The population of normals exhibited a range of mechanisms active in TBU instances. (ii) Individual subjects exhibited different mechanisms in different instances of TBU, even within a single trial. (iii) Individual subjects could in some cases be distinguished from each other based on the distribution of parameters responsible for their TBUs (Fig 1). (iv) Osmolarity increases with increasing evaporation rate at less than a linear rate (Fig 2).
Conclusions :
Intensity decay in TBU areas yielded new data on the mechanisms of TBU in many instances. Quantitative estimates for TBU parameters were variable within and between subjects. The data provides a valuable baseline for the mechanisms and spatial distribution of TBU in normal subjects.
This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.