1,723
Views
2
CrossRef citations to date
0
Altmetric
Cornea

Diagnosis of Dry Eye Disease Using Principal Component Analysis: A Study in Animal Models of the Disease

, , & ORCID Icon
Pages 622-629 | Received 17 Apr 2020, Accepted 22 Sep 2020, Published online: 14 Jan 2021

Figures & data

Figure 1. Histologic changes in inferior lacrimal gland following Concanavalin A

Hematoxylin-eosin stained section of inferior lacrimal gland taken following injection of Concanavalin A. Effacement of normal structure is noted with a severe lymphocytic infiltrate. Only remnants of normal tubuloalveolar structure are visible (yellow arrows); areas of hemorrhagic tissue necrosis with liquefication are seen (red arrow).
Figure 1. Histologic changes in inferior lacrimal gland following Concanavalin A

Table 1. Induction of DED in the three study groups

Figure 2. Principal Component Bi-Plots for surgical model of DED

A: Biplots of PC1 vs PC2 (left) and PC1 vs PC3 (right) provide the best meaningful separation of the data without any overlap between Baseline (normal eyes, grey circles) and DED (black circles). Red triangles show the direction and magnitude for the contribution of active variables used to generate the PC axes. The light blue dashed line represents the mathematically defined cutoff for normal versus DED states.
Figure 2. Principal Component Bi-Plots for surgical model of DED

Table 2. Spearman correlation coefficients between DED parameters one week after its induction

Table 3. Relative contribution of the variables to principal components (%)

Figure 3. Variance for the surgical model of DED. The eigenvalues and explained variation for each of the four principal components (PC) is shown. The red line at the top depicts the cumulative variability explained, with the variability of each successive PC being added to that of the preceding one, reaching 100% with PC4. The majority of total variation (>70%) is explained by the first component

Figure 3. Variance for the surgical model of DED. The eigenvalues and explained variation for each of the four principal components (PC) is shown. The red line at the top depicts the cumulative variability explained, with the variability of each successive PC being added to that of the preceding one, reaching 100% with PC4. The majority of total variation (>70%) is explained by the first component

Figure 4. Principal Component Bi-Plots for the ConA groups

a: Biplots of PC1 vs PC2 (left) and PC1 vs PC3 (right) of the Normal ConA group show complete separation of eyes without overlap between normal (Baseline, grey circles) and DED induced in previously untreated eyes using ConA (black circles). b: Biplots of PC1 vs PC2 (left) and PC1 vs PC3 (right) for the Abnormal ConA group show incomplete separation with overlap between Baseline (grey squares) and DED induced in previously treated eyes using ConA (black squares). Although the baseline group appeared normal based on individual parameters, PCA shows this as not normal, no longer having a linear distribution and now occupying a position further to the left along PC1 (the DED side). In both A and B, the red triangles indicate the direction and magnitude of the active variables that generated the PC axes. The light blue dashed line represents the mathematically defined cutoff for normal versus DED states derived from the surgical model.
Figure 4. Principal Component Bi-Plots for the ConA groups

Table 4. P values for individual clinical test parameters and PC scores

Supplemental material

Supplemental Material

Download Zip (308.6 KB)