RELIABLE CLASSIFICATION OF VISUAL FIELD DEFECTS IN AUTOMATED PERIMETRY USING CLUSTERING

Automated perimetry allows examination of the visual field for diagnostic purposes. Location, shape and size of defects in the visual field detected during a perimetric examination are characteristic hints for the underlying disease of the visual system. Thus a reliable identification of defect types is essential for the proper treatment. We present a classifying system based on cluster analysis and Self-Organizing Maps for the automatic classification of visual field defects. The classifying system distinguishes between eight defect classes and was evaluated on over 8.800 perimetric examinations with a mean classification success of 78%. The classification algorithm is integrated into a software package that can be run on common computers using minor resources; its output can be considered as a suggestion for the physician. As the classification framework is decoupled from the perimetric hardware, it can also be used for the remote classification of perimetric examinations, e.g. in tele-medicine.