The final similarity measure is a sum of these
values over all leaves N. Ordering of the prototypes are by decreasing order of this similarity measure. The accuracy is the percentage of retrieved cases whose label matches the query. The heterogeneous variables include age, gender, postprandial blood glucose (PPBG) levels and family history of the disease (e.g. diabetes). The attribute chosen to separate the population into two homogeneous groups is PPBG, which is less than 180 for normal population. Subsequently, image features are analyzed for classification as shown in Fig. 2. For AMD, drusen (yellowish) and haemorrhages are considered. For DR, reddish spots (MA) or yellowish patches (HE), fluffy white (CWS) or haemorrhages.