Research On The Microaneurysm Detection Algorithm In Digital Fundus Images

The development of signal and information processing technology has bound digital image processing and computer-aided diagnosis ever more tightly together. There are a number of diseases, particularly vascular disease, which leave tell-tale markers in the retina. So, there is much interest in computer analysis of retinal images for identifying and quantifying the effects of diseases such as diabetes. Diabetic retinopathy is a common and severe complication of diabetes, which damages to the retina and even leads to blindness. Therefore, early detection of diabetic retinopathy can help prevent significant vision loss. Since microaneurysms are regarded as the first clinical symptom of diabetic retinopathy, the accurate detection of microaneurysms in retinal images is a critical step for early detection of diabetic retinopathy. This research aims to propose an efficient approach for microaneurysm detection in digital fundus images.This thesis overviews the research background, introduces the research status and significance, and describes the research rub and challenge at first. According to the basic principles of microaneurysm detection, there are three key steps for an efficient microaneurysm detection approach. They are image enhancement, candidate microaneurysms extraction, and screening filtration with features of microaneurysms. Image enhancement can be achieved by using appropriate preprocessing technology, and this thesis focuses on candidate microaneurysms extraction and screening filtration.This thesis mainly researches on microaneurysms extraction using template matching method. A new adaptive template matching approach using dynamic multi-parameter template with the restriction of sum of errors and correlation coefficients is proposed for candidate extraction. The dynamic multi-parameter template matching scheme is more realistic compared to conventional schemes, because it can adjust parameters dynamically to match the target region, namely candidate microaneurysm. So it can match multivariate characteristics of all microaneurysms. After extraction, double lever selection scheme is used to remove any candidate on vessels since microaneurysms cannot occur on blood vessels and reproduce microaneurysms to its real size to remove candidates whose size is considerably big. Experiments indicate that this novel extraction algorithm can effectively improve true positive rate of microaneurysm detection.Finally, candidate microaneurysms achieved by the dynamic multi-parameter template matching algorithm are used for screening filtration based on features of microaneurysms. There are two kinds of strategies are proposed for screening filtration. One is a scoring scheme based on distribution character and the other one is the adaptive weighted summing scheme. The former is used to score every feature value of candidates and the latter is used to calculate weight coefficients for these scores. Then, a total score is calculated for every candidate, and it is used to select true microaneurysms. Experiments have shown that this screening filtration algorithm can decrease false positive rate in microaneurysm detection.To sum up, microaneurysm detection algorithm in this paper can increase detection sensitivity by an average of about 12.2%, and it also decrease the average false positive rate by about 2.14 times while maintain the same true positive rate. Therefore, the proposed algorithm is effective.