Abstract
Purpose :
The aim of this study is to create and demonstrate the usefulness of a new software tool for visualizing and analyzing retinal vessel function analysis data (as assessed by the DVA, Imedos, Jena, Germany) that would enable clinicians to effectively categorize patients in various risk categories based on their measurements.
Methods :
Retinal vessels analysis (RVA) data was collected from 5000 healthy individuals and patients with various CVD risks using a common protocol. Each measurement consists of a series of 350 vessel diameter measurement points, which are treated as 350 dimensional real vectors; each vector is a time series reflecting the change in diameter of a selected vessel segment during the 350s measurement.
Visualizing RVA data can reveal the inherent groupings in the data and could help identifying individuals with similar RVA profiles. While the RVA data of a single patient can be easily plotted, the simultaneous visualization of all measurements from all patients in an understandable way requires dimensionality reduction which maps each RVA measurement to a single point in a two dimensional space. The mapping was performed using t-Distributed Stochastic Neighbor Embedding (tSNE), a state-of-the-art algorithm that is well suited for the visualization of high-dimensional datasets. As an extra layer of information each 2D point is colored according to the corresponding RVA measurement's difference from an age-dependent, averaged, provenly healthy RVA profile.
Results :
Considering the whole dataset, the proposed tool allowed the identification of similar profiles and emerging clusters. For each patient, the tool allowed both a relative positioning in 2D (close to points which denote similar RVA profiles) and an absolute positioning using colors (corresponding to the distance from a predefined standard RVA profile).
Conclusions :
The proposed RVA visualization is a software tool which could help clinical experts categorizing patients and, hence, contribute to a better diagnosis or a more informed follow-up. By identifying how far a patient is from the age-matched healthy control, we could quantify either the degree of risk or the efficiency of a treatment, depending on each case.
This is an abstract that was submitted for the 2018 ARVO Annual Meeting, held in Honolulu, Hawaii, April 29 - May 3, 2018.