OCT INSIGHTS

CLINICAL APPLICATIONS OF ADVANCED IMAGING TECHNOLOGY

The Evolution of Quantitative OCT

BRANDON J. LUJAN, MD

Can OCT move forward as a quantitative technology? The transition to spectral-domain OCT systems has dramatically increased the amount of data obtained, a trend that will continue as systems become increasingly faster and more advanced. The amount of data generated cannot all possibly be reviewed by a busy clinician, and consequently its interpretation depends on a software solution, a segmentation algorithm. However, there is currently no universal algorithm that can be applied to data acquired across different SDOCT systems.

Segmentation algorithms have been an intrinsic component of OCT technology since the original Stratus OCT. While the software of these systems was fraught with errors in the accurate placement of its boundaries, the rules governing its behavior were consistent. The software would look for the only bright line on the back of the scan, which was deemed the “outer retina.” SDOCT has shown us that this line is actually composed of multiple components, and there is no universal consensus as to what should constitute the end of the retina. While it would seem logical that improved resolution would lead to a more precise way to identify boundaries, in actuality the ability to discern more layers has led to more variability.

Without a consensus about where in the outer retina these lines should be placed, different segmentations result in a significant difference of retinal thickness measurements. This is due to the fact that each SDOCT system not only utilizes a unique method of acquisition, but also incorporates an individualized way to analyze these data. This has taken the form of proprietary software algorithms, which serve as a way to distinguish between SDOCT systems. Because these algorithms have presented a significant intellectual hurdle, software is a key selling point for SDOCT systems, with several companies marketing their software as best.

This presents a serious impediment when combining patient data acquired across different platforms; in short, clinical trials suffer. Different software algorithms prevent data from being directly compared across SDOCT systems. These differences in algorithms cannot be accounted for simply by adding in a correction factor; indeed, their variability lies in the basic rules that the algorithm follows and which is accentuated by the presence of pathology. Consequently, clinical trials that aim to use OCT quantitatively are forced to allow only clinical sites that have purchased one particular system. This unfortunately prevents willing clinicians and patients from participating in these trials and thus hinders progress.

No automated segmentation algorithm can be perfect, by definition. Implicit in the algorithm programmer's rules regarding reflectivity changes is a degree of variability that may lead to segmentation failures if the statistical probabilities over what represents a distinct reflectivity profile are offset by unanticipated pathology. Manual segmentation, however, is not a viable solution. The sheer volume of data acquired makes identification of the retinal borders on each of the slices impractical. Even if time weren't a factor, the use of manual graders would introduce significant variability. The axial resolution possible with SDOCT allows outer retinal structures to have a thickness that is subject to interpretation by individual graders as to where lines should be drawn, and there is not necessarily an agreement as to whether the anterior or posterior part of a layer should be segmented. Consequently, reproducibility would be compromised.

The goal of future algorithm development should be one that can be used across different SDOCT platforms and is automated, consistent, and accurate. Freeing data from proprietary algorithms will accelerate the pace at which this goal can be realized. There is precedence for this migration toward “open-source” solutions in radiology, where data are universally analyzable independent of how they were acquired.

In our recent embrace of volumetric imaging technology, ophthalmologists have experienced some necessary growing pains. The initial development of spectral-domain OCT systems has been characterized by a splintered approach to data analysis. It is incumbent upon retinal specialists to direct the evolution of OCT technology toward unified, quantitative outcomes.RP