Abstract
Purpose :
Evaluation of cardinal clinical signs of many ocular surface conditions has historically lacked objective and quantitative data, relying on human observation and subjective grading. The variability produced by the latter has limited the advancement of diagnostic and therapeutic strategies. We evaluated an objective automated measurement system, supported by image acquisition and analysis algorithms, for the quantification of ocular surface pathologies in a clinical study.
Methods :
Patients with dry eye disease (DED) were evaluated with the imaging system in an eye care suite following a standard protocol of corneal fluorescein staining (CFS). CFS was evaluated under the slit-lamp with the NEI scheme. The system performed automated image analysis to output the percentage of the corneal region covered with an epitheliopathy (automated score). Additionally, an experienced clinician labeled each image for regions with epitheliopathy, to determine a manual % score. The automated scores were compared to i) the clinical NEI grading scheme and ii) the manual scores using mixed-effects models. The test-retest repeatability was evaluated using Bland-Altman Limits of Agreement (LOA).
Results :
A total of 208 CFS scans were performed, twice on each eye of 52 patients. 2 eyes were excluded from the analysis as epitheliopathy covered more than 90% of the cornea. The average[SD] NEI score was 7.48[5.7], with 12 eyes graded at NEI=15 and 10 as NEI=0 on at least one measurement. Eyes graded as NEI=15 had a large variation in the manual CFS scores (mean[SD]: 38.3%[20.1], min=16.5%, max=72.6%).
The automated scores were significantly associated with manual scores (β=0.514, 95% CI: 0.45 - 0.59, p < 0.001, R2 = 0.71) and NEI scores (β=0.23, 95% CI: 0.23 - 0.31, p < 0.001, R2 = 0.72). The test-retest LOA for automated scores was ±5.6%, whereas it was 11.7% for the manual scores. For NEI grading, the LOA was ±3.9 points.
Conclusions :
The scores of the severity of epitheliopathy in DED from the automated imaging system were objectively quantified, not human-influenced, and demonstrated a high correlation with scores obtained by clinical experts. The NEI scale, being discrete, fails to capture the wide range of defects with the same granularity as image-based scoring. The performance and scalability capabilities of the system open a frontier for mass evaluation of corneal epitheliopathy with increased repeatability and reliability.
This abstract was presented at the 2023 ARVO Imaging in the Eye Conference, held in New Orleans, LA, April 21-22, 2023.