Optical Coherence Tomography analysis based prediction of Humphrey 24-2 visual field thresholds in patients with glaucoma

Abstract

Purpose : We report a method for prediction of Humphrey 24-2 visual field (HVF 24-2) thresholds from structural analysis of Optical Coherence Tomography (OCT) images. Here, we focus on performance improvements achieved by incorporation of retinal ganglion cell–axonal complex (RGC-AC) concept compared to more conventional approaches.

Methods : We recruited consecutive subjects with early, moderate and advanced primary and secondary open angle glaucoma. Each patient underwent HVF 24-2 testing and 9-field Spectralis SD-OCT imaging. After alignment of 9-field scans, nerve fiber layer (NFL) and ganglion cell and inner plexiform layer (GCL+IPL) were co-segmented and partitioned into 54 sectors matching the HVF test sectors. Subjects with insufficient quality segmentations were excluded from further analysis. Four HVF prediction models were designed and compared: Naive: GCL+IPL and NFL thickness for the predicted sector only; Donut: NFL thickness of 10 sectors that form a Donut centered on the optic nerve head; Garway-Heath: GCL+IPL thickness for the predicted sector and all NFL sectors that fall within the Garway-Heath nerve fiber bundle distribution; RGC-AC optimized: GCL+IPL thickness for the predicted sector and NFL thickness for a set of sectors determined via RGC-AC concept optimization on the training set. The training set consisted of the 9-field OCT and HVF 24-2 thresholds of 111 patients with early, moderate and advanced glaucoma from our previous study [Bogunovic et al, IOVS 2015]. Paired t-tests were used to compare the correlation R between the actual and predicted thresholds for the four models.

Results : We included 86 subjects. Figure 1. shows the correlation R achieved by the four models. RGC-AC optimized performed best prediction with average R=0.74 (95% CI 0.67-0.76), significantly higher than all 3 others (p ≤0.03): Naïve 0.49 (95% 0.44-0.54), Garway-Heath 0.66 (95% 0.60-0.68), Donut 0.67 (95% 0.61-0.69). Figure 2. shows a detailed comparison between actual HVF 24-2 and RGC-AC optimized predicted VF 24-2 thresholds for a specific subject with advanced glaucoma.

Conclusions : The proposed RGC-AC optimized model based on OCT image analysis and the RGC-AC concept is approaching the typical reproducibility of HVF testing.

This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.

 

Figure 1. Sector-Specific predictive performance R for each approach as well as average R over all sectors.