Leung and colleagues have introduced a method for post-processing OCT scans that enhances visualization of axon bundles within the retinal nerve fiber layer when viewed as en-face projection images. Their study also shows that the technique provides improved diagnostic accuracy for detecting glaucoma, as compared to an array of current standard clinical measures. Leung and colleagues have called their new technique: 'Retinal Nerve Fiber Layer Optical Texture Analysis (ROTA)' and their paper describes in detail each of its five post-processing steps, which serve to: (1) reduce variation of RNFL reflectance intensity from B-scan to B-scan within and between eyes by normalizing to the intensity of a reference layer, the RPE; (2) enhance the contrast between axon bundles and the glial-vascular tissue components surrounding the axon bundles within the RNFL by applying an exponential transformation of the normalized pixel intensities; (3) prepare an en-face projection image by summing the normalized, amplified values of RNFL pixels in depth along each A-line; (4) account for RNFL thickness differences across quadrants and eyes by normalizing the en-face image values to a constant, alpha); and (5) further enhance RNFL bundle visibility in the final displayed en-face image by applying another exponential transformation (essentially a gamma function), which helps utilize the full range of available gray scale values for display in an optimal manner.In their paper, Leung et al. show that these post-processing steps produce compelling images of RNFL bundles across the entire posterior pole in healthy eyes and most important, also produce high contrast images of RNFL bundle defects in glaucomatous eyes more reliably than other current standard techniques such as 'red-free' fundus photographs, RNFL thickness maps derived from OCT scans and en-face projection images of RNFL that are not subject to all of the post-processing steps of ROTA. Thus, ROTA helps overcome many of the limitations known to constrain the utility of these other common methods.
Assessment of ROTA images improves diagnostic accuracy over current standards for glaucoma
The study by Leung et al. also clearly demonstrates that assessment of ROTA images improves diagnostic accuracy over current standards for glaucoma, albeit by a modest degree, nevertheless likely to have clinical importance going forward. The methodological details underlying this latter aspect of the study were articulated less clearly than other aspects within the paper, but seem to be based in part by applying a criterion size (minimal two-dimensional area of at least 0.29 mm2) to define an 'RNFL defect' in a ROTA image, as well as some subjective determinations about whether or not a putative defect follows accepted trajectories of RNFL bundles; in any case, there was excellent inter-observer agreement for the latter, which may be further refined in the future to become objective and automated. Similarly, the 'depth' of what defines a ROTA defect was not described in quantitative terms, rather only as being 'hypo-reflective', so, this will also require further scrutiny and verification using other datasets.Naturally, the ROTA technique also provides insight into RNFL bundle loss (and swelling) due to other forms of optic neuropathy besides glaucoma, as shown clearly in this study by Leung and colleagues. Overall, the paper is excellent (including the supplemental material, which provides additional text to describe methodological details, 25 additional figures and 25 additional tables to further support the conclusions); readers interested in the topic are strongly encouraged to peruse the entire report. It should be mentioned that the history of this particular field (starting with the pioneering work of Robert Knighton and colleagues) was not covered well or credited, nor were the many other previous studies about using RNFL reflectance for assessment of glaucoma. Several earlier reports documented similar if not exactly the same approach as some of the post-processing steps used in ROTA. This omission does not undermine the actual results of the study, which were presented very well in this paper. Finally, there are fundamental challenges remaining to overcome for this field, including for ROTA, such as how use of normalization steps based on RPE reflectance (and/or other depths of an OCT scan), which while beneficial in the short run, may introduce problems due to effects of decreased attenuation through degenerative RNFL bundles (i.e., the OCT beam incident on the RPE under an RNFL defect is less attenuated). And, given that nearly all current OCT image segmentation algorithms depend to a great extent on the reflectance differences to define layer boundaries, it may become a circular problem for analysis of RNFL reflectance. In context of these challenges, a novel approach contributed recently to this field by Cannon, Bouma and Uribe-Patarroyo also should be recognized.1