Retinal Vascular Caliber Measurement through Mahalanobis Distance Function based Segmentation

Purpose: To measure retinal vascular caliber by image segmentation with training based on color image segmentation using Mahalanobis distance function. Method: cross-sectional observational study with 20 eyes of 10 normal individuals. Fundus photographs were captured and processed for target feature segmentation using MATLAB R2008 a. Data matrices for all the classes were built followed by calculation of mean and covariance matrix for respective classes. Subsequently, to segment out different target classes from the fundus images, each pixel in the image was mapped to one of the classes giving minimum value of Mahalanobis distance. Optic disc was segmented out. Two circles were drawn at radial distances 0.5D (disc diameter)and 1D from the boundary of the disc. Edge detection was performed through gradient calculation and vascular caliber was determined using two points marked on the walls of a vessel where they intersected the two concentric circles. Result: Average arterial diameter was 147 ±16μm and average venous diameter was 168 ±22μm in our study population. Conclusion: Method used in the present work relies on segmentation by manual marking of the pixels over different study classes to calculate their respective covariance matrices used in Mahalanobis distance function taking care of the correlation of different variables in the feature set. Thus the resulting segmentation determines the edges of the vessels properly which results in accurate computation of vessel width.