Abstract
Purpose:
To examine whether combining Optical Coherence Topography measurements using the ‘Random Forest’ decision tree method improves the diagnosis of glaucoma
Methods:
SD-OCT (Topcon 3D OCT-2000) and perimetry (Humphrey Field Analyzer, SITA standard, 24-2 or 30-2 test pattern) measurements were conducted in 293 eyes of 179 subjects with glaucoma or suspected glaucoma. Visual field (VF) damage was used a ‘gold-standard’ to classify glaucomatous eyes. VF damage was defined as a pattern standard deviation (PSD) value, or, a Glaucoma Hemifield Test (GHT) result, outside normal limits. In total 224 out of 293 eyes (76.5%) had glaucomatous VF damage. The ‘Random Forest’ method was then used to analyze the relationship between the presence/absence of glaucomatous VF damage and the following variables: age, gender, right or left eye, plus 238 different OCT measurements (including axial length). The area under the receiver operating characteristic curve (AROC) was then derived using the probability of glaucoma as suggested by the proportion of votes in this Random Forest classifier. For comparison, four AROCs were derived based on individual OCT thickness measurements of: (i) the macular retinal nerve fiber layer (m-RNFL) alone, (ii) the mean circumpapillary retinal nerve fiber layer (cp-RNFL) alone, (iii) the ganglion cell layer + inner plexiform layer (GCL + IPL) alone, and (iv) rim area alone.
Results:
The AROC from the combined Random Forest classifier (0.90) was significantly larger than the AROCs based on individual measurements of m-RNFL (0.86), cp-RNFL (0.77), GCL + IPL (0.80), and rim area (0.78).
Conclusions:
Evaluating OCT measurements using the Random Forest method provides an accurate diagnosis of glaucoma.
Keywords: 550 imaging/image analysis: clinical •
758 visual fields