Parallel grading of venous beading on transputer

Describes a parallel algorithm and its implementation on transputers for automated grading of venous beading in digitized ocular fundus images. The algorithm mainly consists of median filtering, thresholding, thinning, morphological closing, diameter measurement, fast Fourier transform (FFT) analysis and grading. Median filtering reduces the noise in the original image, thresholding roughly extracts the vein from the background, morphological closing fills holes in the vein, thinning obtains a centerline representation of the vein, diameter measurement and analysis are performed on each centerline branch, and the FFT analyzes the diameter functions. The magnitude spectrum of the diameter function is computed. Usually veins without beading exhibit only low frequency components while beaded veins have significantly more high frequency components. The sum of the high frequency magnitude can be a parameter to distinguish normal from beaded veins. All work has been done on a transputer using PC Trollius