described the various image processing techniques that are available for lamina pore visualisation.

For this project, two cSLO instruments were available to acquire lamina cribrosa images- the Heidelberg Retina Tomograph (HRT) and the Zeiss cSLO. W e have seen from chapter 2 that the two instruments differ fundamentally in their image grabbing techniques, and the methods of processing the images obtained with each are quite different initially (table 2.2). The purpose of this study was to compare and contrast the images obtained with each instrument with view to establishing the more appropriate method for measuring lamina pore

morphology.

3.1.2 Materials and Methods

3.1.2.1 Image Acquisition

In vivo images of the lamina cribrosa were obtained using the Heidelberg Retina Tomograph (HRT, Heidelberg Engineering, Germany) and the prototype Zeiss cSLO (Carl Zeiss, Oberkochen, Germany).

HRT images were acquired using the 15 degree field of view setting. Images were acquired as set out in the HRT manual (Manual 1997) - refractive error and depth settings were set as instructed by the HRT software. The HRT software (version 2.01) has a quality control mechanism that informs the operator whether the images obtained meet the criteria set in the software. A good image set will display a gaussian-like curve of brightness values through the 32 images series, so that the topography of the optic nerve head under investigation is accurately plotted. The HRT displays a comment if the image series is deemed to be of good quality, but the series still has to be inspected for eye movements. Image series with eye movements that caused a distortion in the appearance of the

Image acquisition with the Zeiss oSLO used the HeNe laser light source, with confocal stop size 3 and a 20 degree field of view. The focus setting was placed at the level of sharpest focus of the centre of the optic cup. Images were

recorded onto videotape and then digitised with custom-made digitisation software (Halfyard, Wade et al. 1999) as detailed in section 2.1.1.2.

Images obtained with both instruments were scaled to compensate for image distortions (Zeiss) and magnified to achieve correct image size for Fast Fourier Transformation (Zeiss and HRT) as detailed in section 2.1.

3.1.2.2 Imaae Processing - HRT images

Image series obtained with the HRT were aligned using HRT version 2.01 software, and then converted to individual TIFF format images using the HRTCONV program that is available with the HRT software. The thirty-two individual tiff files underwent a blind deconvolution using AutoDeblur software (version beta 7.5, AutoQuant Imaging Inc., NY, USA). The blind deconvolution served to eliminate the out of focus blur as detailed in section 2.1.4.4.

The result of applying the blind deconvolution is a series of 32 images that have a greater degree of resolution at each section depth. The sections that are of interest in this project are those that are at the depth of the lamina pores. Settings for depth of scan ranged between 2.5mm to 4mm in 0.5 mm steps. Thirty-two optical sections were obtained for each depth setting, giving spacing of between 78 and 125 microns per optical section. In order to make comparisons with the Zeiss images, the optical section that was focused at the level of the neuroretinal rim was used in the analysis. Deconvolved images were taken into Scion Image software (National Institute of Health, Maryland, USA). As with the Zeiss images, F F Ts were generated for the HRT images and custom made filters were applied.

3.1.2.3 Image Processing - Zeiss cSLO

Scion Image software (National Institute of Health, Maryland, USA; shareware) was used to generate the FFT of the averaged lamina image, and a custom

made filter was designed to filter out the unwanted spatial frequency information from the image, as detailed in section 2.1.4.3.

3.1.2.4 Image Quality Measurement: Variance

In order to determine which of the instruments gave the most information about the lamina area, the variance of the intensity of each pixel within the optic disc area of the images was measured. The use of pixel variance in assessing image quality has been described in section 2.1.5. For these images, only the variance within the optic disc area was assessed. However, the brightness of the images varies from image to image, depending on the instrument used as well as the individual subject (e.g. due to lens opacities), and, in addition the optic disc occupies different areas in the HRT and Zeiss images. These differences would certainly affect the ‘spread’ of intensity values in both the HRT and Zeiss

images, thus causing a difference in the variance measure that would not be due to pore detail detected alone.

The use of the ‘histogram arithmetic’ function in the image processing technique described earlier (section 2.1.4.6) would help counteract the effect of the

difference in brightness, or luminance, of the images. All the processed images are shifted so that they have a mean luminance of 128, thus eliminating the influence of brightness difference on the spread of intensity values.

The variance formula uses the number of pixels in the data set in its calculation. As the optic disc occupies a smaller area in the HRT image compared with the Zeiss image, the number of pixels measured in these two images will be different for each subject (figure 3.1). This will therefore affect the variance measure. A method of overcoming this would be to count the number of pixels in each data set, and take a ratio of the variance/number of pixels. This ratio was termed ‘variance per pixel’ (VPP)- a misnomer, because in fact there would be no such thing as variance per pixel per se. However, for the purpose of this thesis, it was felt that the term was suitable.