OCT Angiography Technologies
OHSU, Casey Eye Institute, July 25th, 2015.
This past weekend I attended the first OCT angiography summit at the Casey Eye Institute, at Oregon Health and Science University. Held at an extremely impressive facility, this inaugural summit brought together the key innovators in OCT angiography (OCT-A), both from a clinical and technological perspective. In the following I post my notes from the OCT angiography technologies session.
This session featured two of the original OCT inventors, one of whom, Dr. David Huang, gave the introductory talk. Entitled, “Split-spectrum amplitude decorrelation angiography”, the talk rightly introduced OCT angiography as the future of OCT, a functional imaging technique based on motion contrast. Following a brief overview of the main pioneers in this area, the talk concentrated on the Split-Spectrum Amplitude Decorrelation Angiography Algorithm (SSADA) that was developed at OHU by Dr.’s Jia and Huang . Based on OCT imaging in the same location, decorrelation values across pixels is an indication of blood flow; correlation being an indication of static tissue. The technique uses intensity only information – i.e., not phase – and cleverly avoids bulk motion artifact by exploiting the fact that the motion of interest is transverse. This is done by effectively reducing the axial resolution, where such bulk motion dominates, by band-pass filtering in the Fourier domain. This is done over multiple bands, however, as the all important speckle signal is dependent on wavelength, and, therefore, each resultant spectral band contains unique information that is not smeared out were only single band used. This stack of image data is then recombined by averaging, which further improves the overall signal of the resulting angiogram. As was emphasized during the clinical sessions, this is wonderful technology.
This introductory talk set well the tone of the entire symposium as the SSADA technology has been licensed to Optovue Inc., who use it in their Avanti AngioVue product. Being the first commercially available OCT-A system, this was naturally the device used in the vast majority of the clinical talks given throughout the day. Indeed, the next talk was about this device, given by the VP of R&D at Optovue, Dr. Tony Ko. It has some incredibly clever engineering, two pieces of which are licensed from academia: SSADA from OHSU and Motion Correction Technology (MCT) from MIT/Erlangen. Dr. Ko gave an overview of these (more of which I will cover layer in the talk by Professor Fujimoto), their advantages and then what scan patterns the device supports. Engineering improvements of SSADA are given in , and internal optimizations of the MCT algorithm reduce computation time from ~5 minutes to ~10 seconds using GPUs. Importantly, full volumetric OCT-A data can be acquired at the macula or optic disc in around 3 seconds. This, in part, is because the motion correction is done as a post-processing step, which has an inherent advantage to the patient who need not spend a long time on the chin rest; the disadvantage to real-time tracking (see TruTrackTM and FastTrac) is that it can take a long time as the data acquired is only done so when in the known, correct position. Automatic segmentation delineates the circulatory layers of interest, allowing for visualization of the superficial and deep capillary in the macula.
Dr. Yan Li, from the Casey Eye Institute, presented “OCT Angiography of the Cornea and Iris”, which was also an initial introduction to the Optical microangiography (OMAG) technology . Imaging on human patients was done using the Avanti system from Optovue as well as a high-speed anterior segment swept-source OCT (SS-OCT) system, developed in collaboration with MIT. The former system is spectral domain OCT (SD-OCT), using an 840nm light source; the latter uses a 1050nm tunable laser. The SS-OCT system captures 100,000 axial scans (a-scans) per second. Both were shown to have clinical value in imaging corneal neovascularization, melanoma and iris nevus, but the longer wavelength of the SS-OCT device was more useful for tumors, where depth penetration was important to overcome pigmentation.
Dr. Huang’s SSADA co-inventor, Dr. Yali Jia, gave the next talk, “Advanced image processing for OCT angiography”. Skipping ahead a little, it’s worth noting that the methods discussed were being used on clinical data presented in other talks when, truly, more advanced, custom methods were required. This talk introduced the flow projection artifact that can occur in OCT-A, when blood vessels in, for example, the retinal nerve fiber layer (RNFL) interfere with light reaching the retinal pigment epithelium (RPE) layer and present as high signal there in the OCT-A image. As such, the blood flow in the superficial vascular bed is resulting in OCT signal in the deeper layers. In the normal retina the imaged flow is artifact. Correction for this is necessary, as when there is signal it could be a sign of neovascularization. The correction, in short, is based on a vessel mask, derived from the superficial retina. When signal is then seen, this can indeed show as areas of choroidal neovascularization (CNV), which can clearly be visualized in the en face view. It’s hard to delineate the extent of the area, however, but certainly of clinical interest. Using morphological processing, the area was well delineated and resolved in the data show. Another example was in diabetic retinopathy (DR), where layer segmentation is the important step to limiting the integration range used of the OCT-A data in order to visualize the appropriate vascular network. For this work they used graph-traversal techniques that have been widely adopted for OCT segmentation, both commercially and in academia. A limitation is that they are inherently 2d segmentation techniques, and less powerful than the fully 3d graph-cut methods, but a lot faster. Consequently, Dr. Jia was only able to show results of 2d segmentations, but those results looked very good.
Up to this point we had only seen data and applications based on OCT-A using SSADA. This, as noted, is solely an image intensity based method for visualizing flow, and does not use phase information. The talk by Professor Ruikang Wang of the University of Washington, a pioneer in this area, showed the use of phase in addition to image intensities in the generation of data. This technology is licensed to Carl Zeiss Meditec and is set to debut at AAO’s annual meeting in November of this year. Two devices will come out, an SD-OCT using an 840nm laser and capturing data at 68K a-scans/second, and an SS-OCT with a 1060nm light source, running at 100K a-scans/second. Of note, the SS-OCT system does not have motion correction (here, in the form of real-time tracking). The use of tracking does allow for wide-field image capture (>50 degrees), but the downside being the time to acquire the data, although the images in the normal retina did look great. Emphasis was given to the fact that the segmentation algorithms used in forming the angiograms are validated methods. Used are segmentations of the ILM, the posterior of the IPL, the posterior of the OPL and the RPE. A comparison of these algorithms to our own can be found here.
Dr. Andreas Pollreisz, of the Medical University in Vienna, presented a comparative study between the AngioVue and a wide-field bi-directional OCT angiography system developed in conjunction with Professor Leitgeb’s lab. Clever interleaving – “spectral splitting” – of the raster allows for a two-fold increase in data acquisition, resulting in data rates of 200k a-scans/second. In cases of choroidal neovascularization and capillary non-perfusion in diabetic retinopathy, Dr. Pollreisz was able to demonstrate clinically similar flow information between the two different devices.
The final talk of this packed initial session was from the honorary guest, Professor James Fujimoto of MIT. Dr. Fujimoto introduced their SS-OCT system, using very advanced vertical-cavity surface-emitting laser (VCSEL) technology developed by MIT in conjunction with industrial partners. With a wavelength of 1050nm the effect of scattering is reduced, resulting in a higher overall signal to noise ratio. Furthermore, the data can be driven to 400K a-scans/second! Some wonderful images of the choriocapillaris were shown. With such high data rates, patient motion is to some extent reduced, but correction is still necessary. This led to an introduction to the MCR algorithm originally introduced by Dr. Ko of Optovue. Image data is acquired by rastering in orthogonal directions, and, under the assumption that a single B-scan is motion free, a reshuffling of the data utilizing the fact that the fast acquisition is done in both directions, allows for a full correction, both laterally and axially. In essence, this is a constrained but complex optimization problem, where the objective function is the displacement field for the volume . It’s very clever engineering, as although data is acquired in two directions, this repeat data is anyway necessary for angiography, which requires multiple acquisitions in the same location. Furthermore, it is done, quickly, off-line, so it doesn’t require the patient to spend time on the chin rest waiting for a motion free acquisition. And lastly, it reduces equipment cost as a there is no need for an additional camera (line scanning ophthalmoscope) to facilitate tracking. This engineering solution came about in collaboration with Professor Joachim Hornegger’s group at the University of Erlangen, and is a good example of how research labs with a background in image processing in radiology can contribute well to ophthalmic imaging.
To close this extremely interesting session, a short question and answer period followed. A question to Dr.’s Huang and Fujimoto regarding the pros and cons of SD-OCT and SS-OCT angiography was interesting. Despite the obvious speed advantages, Dr. Huang cited lower decorrelation noise and better transverse resolution (driven by the focal spot size, which is already ~2-3 times the size of the capillary diameter) as advantages of the SD-OCT systems. Dr. Fujimoto agreed, but thought the noise characteristics are similar when designed right, but conceded the major disadvantage of SS-OCT being cost. Dr. Wang offered the fact that the choice depends on the optical properties of the tissue being imaged; i.e., retinal tissue is better imaged with a shorter wavelength, whereas the choroid requires a longer wavelength for the improved depth penetration.