Automated analysis of optic nerve head changes in patients with glaucoma

Abstract

Purpose : Glaucoma is a main cause of irreversible vision loss in developed countries. Due to the slowly progressive nature of the disease, controversy in diagnostic criteria, and subjective interpretation of clinical findings, up to 50% of patients remaining undiagnosed. Yet, early detection and intervention are pivotal to delay or prevent functional deficits. This study aimed to develop a windows-based application with the potential to automatically highlight changes around the optic nerve head over time.

Methods : We developed a C# module based software application enabling manual and automated alignment and subsequent detection of significant differences in pixel brightness between individual images. Data was collected from 100 patients who attended at least 5 complete glaucoma assessments over a minimum of 18 months. Red-free and stereoscopic fundus images at baseline and last follow-up were inspected by two ophthalmologists with glaucoma specialty and graded for progression. The same images were analysed using the newly developed software. Levels of agreement between graders and the automated detection outcomes were calculated using Cohen’s Kappa.

Results : During the monitoring period of 3.2 ± 0.8 years, 13 patients were identified with progressive changes by at least one ophthalmologist based on optic disc appearance at the first and last visit alone, resulting in overall fair inter-grader agreement (K=0.24). Using the newly developed software, agreement with each of the observers was fair to moderate (K=0.25 and K=0.48 respectively). The application was most successful at identifying change with Drance haemorrhages or peripapillary atrophy. 12% of optic discs could not be aligned using the fully automated program function only.

Conclusions : According to the World Glaucoma Association consensus, “detecting progressive glaucomatous RNFL thinning and neuroretinal rim narrowing are the best currently available gold standards for glaucoma diagnosis”. Preliminary evaluation of an automated application revealed the ability to highlight relevant clinical changes and aid in such assessment. While not sufficient in isolation, integration of such tools in clinical practice should improve interobserver agreement. Further studies are currently underway.

This is an abstract that was submitted for the 2018 ARVO Annual Meeting, held in Honolulu, Hawaii, April 29 - May 3, 2018.