Purpose:
Corneal epithelium images can be obtained through HeidelbergRetina Tomograph corneal module. Automated analysis of epitheliumcells morphology could help clinical and pathological diagnosis.First step toward this analysis is to extract cell borders froman HRT image. We propose an algorithm to automatically performsegmentation of these images (Fig. 1).
Methods:
Steps of the algorithm include:• A noise reduction througha fast-Fourier-transform (FFT) band pass filter with normalization(pixel values range from 0 to 255).• Image binarizationwith a median value threshold.• Standard watershed algorithmon distance map is used to separate fused cells (Fig 2).• Voronoidiagram segmentation algorithm gives a final border map (Fig3).
Results:
Most cells borders are correctly found. However, a small numberof oversegmented small cells are present.
Conclusions:
Epithelial cell image segmentation can be performed automatically,thus allowing a statistical analysis of cell morphological andpathological characteristics
Keywords: cornea: epithelium • image processing • imaging/image analysis: non-clinical