Applications of Artificial Intelligence and Deep Learning in Glaucoma.

Diagnosis and detection of progression of glaucoma remains challenging. Artificial intelligence-based tools have the potential to improve and standardize the assessment of glaucoma but development of these algorithms is difficult given the multimodal and variable nature of the diagnosis. Currently, most algorithms are focused on a single imaging modality, specifically screening and diagnosis based on fundus photos or optical coherence tomography images. Use of anterior segment optical coherence tomography and goniophotographs is limited. The majority of algorithms designed for disease progression prediction are based on visual fields. No studies in our literature search assessed the use of artificial intelligence for treatment response prediction and no studies conducted prospective testing of their algorithms. Additional challenges to the development of artificial intelligence-based tools include scarcity of data and a lack of consensus in diagnostic criteria. Although research in the use of artificial intelligence for glaucoma is promising, additional work is needed to develop clinically usable tools.

A Region-of-Interest Approach for Detecting Progression of Glaucomatous Damage With Optical Coherence Tomography.

IMPORTANCE: Detecting progression of glaucomatous damage is often challenging. OBJECTIVE: To test the feasibility of using frequency-domain optical coherence tomography (FD-OCT) and a region-of-interest (ROI) approach to measure progressive changes in glaucomatous damage. DESIGN, SETTING, AND PARTICIPANTS: Among a group of patients in an institutional glaucoma practice who were likely to show glaucoma progression, eyes with a history of an optic disc hemorrhage (DH) confirmed by stereophotography were followed up with FD-OCT cube scans of the optic disc. All patients underwent FD-OCT scans on at least 2 occasions separated by at least 1 year (mean, 3.45 years; range, 1.42-6.39 years). Because we were not studying the effects of an optic DH, no constraint was placed on the time between the documentation of an optic DH and the first scan used in the analysis. MAIN OUTCOMES AND MEASURES: After en face images of the FD-OCT scan were aligned based on the blood vessels, circumpapillary images were derived for an annulus 100 µm in width, and the retinal nerve fiber layer (RNFL) thickness profiles were plotted for the first and last visits. The ROI width associated with the optic DH was defined as the region of the RNFL profile below the 1% CI based on healthy norms. The change in the ROI width was compared with the change in the global RNFL thickness, which was obtained by averaging the circumpapillary RNFL thickness. RESULTS: The change in the ROI width (mean [SD], 8.0° [6.4°]; 95% CI, 4.9° to 11.1°; range, -0.7° to 19.3°) was significant (P < .001, 2-tailed t test) while the change in the global thickness (mean [SD], 2.40 [5.87] µm; 95% CI, -0.48 to 5.28 µm) was not significant (P > .12, 2-tailed t test). Although 15 of the 16 ROIs increased in width between visits, only 11 showed a decrease in the global RNFL thickness. CONCLUSIONS AND RELEVANCE: For detecting progression of local RNFL damage in patients with glaucoma, an OCT ROI approach appears superior to the OCT global RNFL thickness measure typically used.

Autofluorescence imaging and spectral-domain optical coherence tomography in incomplete congenital stationary night blindness and comparison with retinitis pigmentosa.

PURPOSE: To test the hypothesis that the evaluation of retinal structure can have diagnostic value in differentiating between incomplete congenital stationary night blindness (CSNB2) and retinitis pigmentosa (RP). To compare retinal thickness differences between patients with CSNB2 and myopic controls. DESIGN: Prospective cross-sectional study. METHODS: Ten eyes of 5 patients diagnosed with CSNB2 (4 X-linked recessive, 1 autosomal recessive) and 6 eyes of 3 patients with RP (2 autosomal dominant, 1 autosomal recessive) were evaluated with spectral-domain optical coherence tomography (SD OCT) and fundus autofluorescence (FAF). Diagnoses of CSNB2 and RP were confirmed by full-field electroretinography (ERG). Manual segmentation of retinal layers, aided by a computer program, was performed by 2 professional segmenters on SD OCT images of all CSNB2 patients and 4 age-similar, normal myopic controls. Seven patients were screened for mutations with congenital stationary night blindness and RP genotyping arrays. RESULTS: Patients with CSNB2 had specific findings on SD OCT and FAF that were distinct from those found in RP. CSNB2 patients showed qualitatively normal SD OCT results with preserved photoreceptor inner segment/outer segment junction, whereas this junction was lost in RP patients. In addition, CSNB2 patients had normal FAF images, whereas patients with RP demonstrated a ring of increased autofluorescence around the macula. On SD OCT segmentation, the inner and outer retinal layers of both X-linked recessive and autosomal recessive CSNB2 patients were thinner compared with those of normal myopic controls, with means generally outside of normal 95% confidence intervals. The only layers that demonstrated similar thickness between CSNB2 patients and the controls were the retinal nerve fiber layer and, temporal to the fovea, the combined outer segment layer and retinal pigment epithelium. A proband and his 2 affected brothers from a family segregating X-linked recessive CSNB2 had a mutation, p.R614X, in the gene encoding calcium channel, α 1F subunit. CONCLUSIONS: CSNB2 patients (X-linked recessive and autosomal recessive) had significantly thinner retinas than myopic controls. However, they demonstrated qualitatively normal SD OCT and FAF images, and therefore can be differentiated from RP patients with these techniques. Although ERG testing remains the gold standard for the diagnosis of these conditions, FAF and SD OCT systems are more widely available to community ophthalmologists, offer shorter acquisition times, and, unlike ERG, can be performed on the same day as the initial clinic visit. Therefore, as a supplement to ERG and genetic testing, we advocate the use of FAF and SD OCT in the examination of patients with CSNB2 and RP.