Overlay of conventional angiographic and en-face OCT images enhances their interpretation

The combined use of morphological and functional information has been shown to improve diagnostic accuracy in many clinical disciplines [1‐3]. In the management of patients with malignancies, coronary heart disease and diseases of the brain, fusion of different image modalities, such as CT, MRI, PET and various other nuclear diagnostic tests, is widely used [3‐7]. Retinal diseases are also very complex, and often more than one diagnostic or imaging technique is used to assist in making a diagnosis. Cunha-Vaz and co-workers [8, 9] have shown the value of multimodal mapping systems for the macula, incorporating retinal imaging techniques such as confocal scanning laser ophthalmoscopy (SLO), the retinal leakage analyzer, the retinal thickness analyzer, automated perimetry and SLO angiography. However, inherent difficulties exist in the ability to correlate these various modalities, as these techniques lack common, precise reference points.

Recently, a new modality in ocular imaging was introduced combining high depth resolution OCT with high transversal resolution confocal ophthalmoscopy [10‐13]. The prototype OCT-Ophthalmoscope simultaneously produces en-face OCT scans and pixel-to-pixel corresponding confocal images, so called OCT C-scans. The OCT C-scans are oriented in a transversal plane, perpendicular to the optical axis of the eye. A stack of such OCT C-scans allows for a reconstruction of the retina in three dimensions. Although this transversal plane is the more familiar plane when viewing the retina, compared to the conventional, longitudinal OCT images [14‐16], single en-face OCT scans may look confusing to the uninitiated, due to their high depth resolution (about 10 micron) and their transversal orientation through the retina [12, 17].

The confocal channel creates a fundoscopic image with a high transversal resolution but a low depth resolution (~300 μm). There is seemingly little change in the confocal image when scanning through the retina along the Z-axis (depth). Therefore the confocal part of the OCT C-scan provides a high quality fundus image throughout the whole OCT stack in depth. As both the confocal and OCT image are acquired simultaneously with the same light source and at the same scanning rate, there is pixel-to-pixel correspondence between the OCT and the confocal image of the OCT C-scan [11]. The confocal image is used for general orientation and localization, whereas the OCT image shows detailed morphology, and subsequent pathological changes within the retina [10].

At the Department of Ophthalmology of the Academic Medical Centre (University of Amsterdam, Amsterdam, The Netherlands) a prototype OCT-Ophthalmoscope (Ophthalmic Technologies Inc., Toronto, Canada) was used to evaluate patients with various retinal pathologies. Patients undergoing routine fluorescein angiography (FA), using Imagenet (Topcon, USA), were scanned with the OCT-Ophthalmoscope in the area of interest on the same day. To demonstrate the possibilities of this new technique, we selected some illustrative, though not necessarily common, pathological cases.

The configuration of this OCT prototype has been described previously [10, 13]. In short, the system uses a super luminescent diode, with a central wavelength of 820 nm (20 nm bandwidth). The light beam is split, directing one part to the patient's eye (sample arm) and the other part to a reference arm (mirror). The returning light beams from both the patient's eye and the reference arm are collected through an interferometer to produce the OCT signal [11]. A fraction of the light returning from the patients' eye is also directed towards another detector to produce a confocal signal [11]. The OCT-Ophthalmoscope produces a transversal OCT C-scan (size: 1042 × 512 pixels), currently at 2 frames/second, in the X-Y-plane at a fixed Z-coordinate, in which the OCT and confocal images are in pixel-to-pixel correspondence with each other [17].

ImagePro is an image-processing program working under Windows^©. By using a custom made plug-in for this program, it is possible to spatially transform images, and to superimpose them over reference images. For the purpose of this study, the confocal part (size: 512 × 512 pixels) of the OCT C-scan was used as a reference image. Since every transformation inherently leads to loss of anatomical resolution in the transformed image, we chose to leave the OCT C-scan intact. The OCT C-scan covers a smaller area than the FA. Detailed information in the OCT scan would be lost due to a necessary volume transformation if the OCT C-scan would be transformed onto the FA. One or two OCT C-scans were chosen from the stack, depending on the depth at which the pathologies were best visible. The appropriate angiographic Imagenet frame was transformed onto, and superimposed over the confocal image. Alignment was then achieved by using evident landmarks as reference points. Inherent to the scanning procedure of the OCT-Ophthalmoscope, using a curved fan ray equivalent to the curvature of the eye, there is some distortion at the borders of both parts of OCT C-scan relative to the centre of the images. We found that it was best to choose reference points centrally within the vascular arcades. Four to six pairs of reference points were manually selected. This was done by zooming in on the area in which a landmark was localised. Landmarks were chosen at vascular crossings or vessel bifurcations that were clearly visible in both image modalities. These reference points were used to compute a linear transformation by the least-square method, compensating for translation, rotation, scaling, and shearing (full affine). The spatially transformed angiographic image was doubled and then superimposed over both the confocal image and the OCT scan. To further facilitate visualization, the images were combined into the different layers of an RGB image: the green channel of the RGB image was used for the spatially transformed angiographic image, and the other channels were used for both parts of the OCT C-scan. Images presented in this paper show on the left-hand side the confocal image, and on the right-hand side the OCT image with or without angiographic overlay. Figure 1 shows the above described procedure for making an overlay image with the help of OCT C-scan and FA images of a normal fundus. The whole procedure (including acquisition and selection of the appropriate image frames) took no more than 15 minutes per patient.

The OCT-Ophthalmoscope provides clinicians with a novel imaging modality of the retina. The transversally oriented OCT C-scans allow software-assisted overlay with more traditional imaging techniques, such as FA or indocyanine green (ICG) angiography, that are also displayed in a transversal plane. The confocal channel in the OCT-Ophthalmoscope provides a fundoscopic image with a high transversal resolution which has a pixel-to-pixel correspondence with the OCT channel that possesses a very high depth resolution [10]. Therefore, the confocal image can easily be used as a reference image. Images produced by other diagnostic techniques can be mathematically transformed to fit snugly over the confocal image of the OCT C-scan, and thereby be directly superimposed over the OCT image.

Our initial goal in making overlay images, was to enhance the interpretation of individual OCT C-scans. These are initially difficult to interpret. The high transversal resolution makes recognition of retinal landmarks difficult in the OCT C-scan images. Classically, pathologic processes are studied in longitudinal sections, so that their transversal appearance is rarely appreciated. By using the overlay technique, it is possible to correlate pathologic features imaged with familiar transversal imaging techniques, such FA, with the morphological information provided on OCT C-scans. We also found that the overlay technique provided additional insight not readily available with either modality alone.

In our example of Birdshot retinochoroiditis (figure 2) the OCT C-scan showed an irregularly shaped retinal contour in the parafoveolar area, and along the arcades. On the FA image diffuse parafoveolar oedema was present as well as diffuse leakage along the major retinal vessels. Leakage correlated well in the overlay to the irregularly shaped retinal contour on the OCT. Thus, the patchy retinal thickening seen on OCT was caused by diffuse oedema, which if needed could be followed prospectively, without the need of the more invasive angiographic technique. In figure 3, OCT and FA concurred in the location of the neovascular membrane. Further, the OCT also showed that the neovascular membrane was situated below the RPE, making this membrane ineligible for current surgical procedures [20], while the retina above was intact and not involved in the process. Combining findings on the longitudinal and transversal OCT allowed one to better comprehend the FA manifestations observed at the edge of the neovascular lesion. Hypofluorescence at the edge of the PED is due to superposition of pigmented RPE cells, while the more peripheral hyperfluorescence is caused by a window defect. The OCT and FA images in the RAP example (figure 4) revealed the limitations of FA in defining the extent of involvement, in particular of the RPE [21, 22]. FA leakage was present in only a portion of the PED, probably due to the presence of blocked fluorescence, although the cause of this blocked fluorescence could not be ascertained with the OCT.

Conventional OCT is a valuable tool in making a diagnosis and in monitoring the progress of macular diseases. The commercially available StratusOCT scans in a longitudinal orientation, providing high resolution, morphologic, cross-sectional OCT scans of the retina, but it does not provide a precise localization of the scanned lines within the macular region, because there is always a time delay between acquisition of the line and storage of the line with its associated fundus image. A retinal thickness map is based on the interpolation of the thickness measurements from six radial OCT scan lines. While these maps are useful to monitor patients, they cannot be combined with other transversal diagnostic techniques because both the map and the scan lines which it is made of, lack good reference points. For this and other reasons, conventional OCT has not yet been incorporated in multimodal mapping systems such as the one developed by Cunha-Vaz and co-workers [8].

Rosen and co-workers [23] were able to produce simultaneous ICG angiography and en-face OCT in an adjusted, experimental OCT-Ophthalmoscope. Here pixel-to-pixel correspondence was seen in both channels, which ultimately may be a better set-up than the technique proposed here. However, this combination will need further optimization before it becomes clinically applicable. At this moment, our examples suggest that multimodal mapping is possible between the OCT-Ophthalmoscope and existing imaging modalities, without various imaging modalities being incorporated into the system. Combining several imaging techniques provides complementary information which helps in interpreting and understanding the findings of each modality. Additional modalities which could possibly be fused with transversal OCT include microperimetry and (multifocal) electroretinography. These would provide functional correlations to structural alterations in the retina.