Purpose:
Generation of dense statistical deformation models (SDM) capturing the optic nerve head (ONH) variability to indicate relevant differences between healthy and glaucomatous cases from color fundus images.
Methods:
While ONH variability is commonly captured by sparse geometric measurements, we represent it by dense deformation fields between one single reference image and images of a sample set calculated by non-rigid image registration. The reference image is not a simple average, but is indicated by the minimal residual deformations.For both, glaucomatous (N=90) and controls (N=90), a SDM is generated by Principal Component Analysis from the calculated residual deformations showing the major variation modes of the ONH. The gold standard diagnosis was given by a glaucoma specialist based on an elaborate ophthalmological examination with ophthalmoscopy, visual field, IOP, FDT, and HRT II.
Results:
Similar to the input images, the constructed reference of the ONH is naturally shaped and is not smoothed as it is the case for simple averaging. Both SDMs show major optic nerve variations such as varying ONH or excavation area (Figure 1 left).Eliminating the control variation modes from glaucoma SDM, the glaucomatous variations exclusively remain. Figure 1 (right) exemplary indicates glaucomatous rim changes in the nasal sector (9 o'clock) of the ONH.
Conclusions:
The proposed SDM approach allows a dense and detailed representation of common and exclusive glaucomatous ONH variations. We expect that these models provide new insights to the glaucoma disease and can be potentially applied for automated glaucoma detection.
Keywords: optic nerve • image processing • shape, form, contour, object perception