Early detection of microaneurysms (MAs), the first sign of Diabetic Retinopathy (DR), is an essential first step in automated detection of DR to prevent vision loss and blindness. This study presents a novel and different algorithm for automatic detection of MAs in fluorescein angiography (FA) fundus images, based on Radon transform (RT) and multi-overlapping windows. This project addresses a novel method, in detection of retinal land marks and lesions to diagnose the DR. At the first step, optic nerve head (ONH) was detected and masked. In preprocessing stage, top-hat transformation and averaging filter were applied to remove the background. In main processing section, firstly, we divided the whole preprocessed image into sub-images and then segmented and masked the vascular tree by applying RT in each sub-image. After detecting and masking retinal vessels and ONH, MAs were detected and numbered by using RT and appropriated thresholding. The results of the proposed method were evaluated reported on three different retinal images databases, the Mashhad Database with 120 FA fundus images, Second Local Database from Tehran with 50 FA retinal images and a part of Retinopathy Online Challenge (ROC) database with 22images. Automated DR detection demonstrated a sensitivity and specificity of 94% and 75% for Mashhad database and 100% and 70% for the Second Local Database respectively.

Detection of red lesions in color retinal images is a critical step to prevent the development of vision loss and blindness associated with diabetic retinopathy (DR). Microaneurysms (MAs) are the most frequently observed and are usually the first lesions to appear as a consequence of DR. Therefore, their detection is necessary for mass screening of DR. However, detecting these lesions is a challenging task because of the low image contrast, and the wide variation of imaging conditions. Recently, the emergence of computer-aided diagnosis systems offers promising approaches to detect these lesions for diagnostic purposes. In this paper we focus on developing unsupervised and supervised techniques to cope intelligently with the MAs detection problem. In the first step, the retinal images are preprocessed to remove background variation in order to achieve a high level of accuracy in the detection. In the main processing step, important landmarks such as the optic nerve head and retinal vessels are detected and masked using the Radon transform (RT) and multi-overlapping windows. Finally, the MAs are detected and numbered by using a combination of RT and a supervised support vector machine classifier. The method was tested on three publicly available datasets and a local database comprising a total of 749 images. Detection performance was evaluated using sensitivity, specificity, and FROC analysis. From the image analysis viewpoint, DR was detected with a sensitivity of 100% and a specificity of 93% on average across all of these databases. Moreover, from lesion-based analysis the proposed approach detected the MAs with sensitivity of 95.7% with an average of 7 false positives per image. These results compare favourably with the best of the published results to date.

Retinal blood vessel segmentation and analysis is critical for the computer‐aided diagnosis of different diseases such as diabetic retinopathy. This study presents an automated unsupervised method for segmenting the retinal vasculature based on hybrid methods. The algorithm initially applies a preprocessing step using morphological operators to enhance the vessel tree structure against a non‐uniform image background. The main processing applies the Radon transform to overlapping windows, followed by vessel validation, vessel refinement and vessel reconstruction to achieve the final segmentation. The method was tested on three publicly available datasets and a local database comprising a total of 188 images. Segmentation performance was evaluated using three measures: accuracy, receiver operating characteristic (ROC) analysis, and the structural similarity index. ROC analysis resulted in area under curve values of 97.39%, 97.01%, and 97.12%, for the DRIVE, STARE, and CHASE‐DB1, respectively. Also, the results of accuracy were 0.9688, 0.9646, and 0.9475 for the same datasets. Finally, the average values of structural similarity index were computed for all four datasets, with average values of 0.9650 (DRIVE), 0.9641 (STARE), and 0.9625 (CHASE‐DB1). These results compare with the best published results to date, exceeding their performance for several of the datasets; similar performance is found using accuracy.

No abstract