Automatic image analysis has much to offer to diabetic retinopathy screening programmes by indefatigably performing disease detection in thousands of retinal photographs. Blot haemorrhages (BH) are one of the signs of diabetic retinopathy and, if present in the macular region or in either hemifield in sufficient number (≥4), are one of the signs that the patient requires referral to an eye clinic. This paper describes an automated system for detection of BH in retinal images. A multi-scale, morphological technique is used for segmentation of BH candidates which are then classified. Images are detected with sensitivity and specificity respectively, 98.6% and 95.5% for ≥4 BH in both hemifields, 91.6% and 93.9% for ≥4 BH in either hemifield, 80.3% and 84.0% for ≥1 macular BH. If combined with automated systems for detection of other signs of diabetic retinopathy, this system could form part of a system for detection of diabetic retinopathy requiring referal to an eye clinic.
