The estimation of Arterio-Venous ratio (AVR) is an important phase in diagnosing various vascular diseases e.g. Diabetic Retinopathy. For calculating this value, it is essential to differentiate the vessels into arter...The estimation of Arterio-Venous ratio (AVR) is an important phase in diagnosing various vascular diseases e.g. Diabetic Retinopathy. For calculating this value, it is essential to differentiate the vessels into arteries and veins. This paper presents a novel structural and automated method for artery/vein vessels classification in retinal images. Our method is tested on DRIVE database and the classification accuracy is 88.7 % for pixels and 89.07 % for vessel lines, respectively, which demonstrate the effectives of our approach. Our method will help to achieve the fundus disease surveillance on mobile and remote medical treatment. It has a remarkable social significance.展开更多
Reliable detection of fundus lesion is important for automated screening of diabetic retinopathy. This paper presents a novel method to detect the fundus lesion in retinal fundus image based on a visual attention mode...Reliable detection of fundus lesion is important for automated screening of diabetic retinopathy. This paper presents a novel method to detect the fundus lesion in retinal fundus image based on a visual attention model. The proposed method intends to model the visual attention mechanism of ophthalmologists during observing fundus images. That is, the abnormal structures, such as the dark and bright lesions in the image, usually attract the most attention of experts, however, the normal structures, such as optic disc and vessels, have been usually selectively ignored. To measure the visual attention for abnormal and normal areas, the incremental coding length is computed in local and global manner respectively. The final saliency map of fundus lesion is a fusion of attention maps computed for the abnormal and normal areas. Experimental results conducted on the publicly DiaRetDB1 dataset show that the proposed method achieved a sensitivity of 0.71 at a specificity of 0.82 and an AUC of 0.76 for fundus lesion detection, and achieved an accuracy of 100% for normal area (optic disc) detection. The proposed method can assist the ophthalmologists in the inspection of fundus lesion.展开更多
基金This work was supported in part by the Natural Science Foundation of China under Grant 61472102, in part by the Fundamental Research Funds for the Central Universities under Grant HIT.NSRIF.2013091, and in part by the Humanity and Social Science Youth foundation of Ministry of Education of China under Grant 14YJC760001.
文摘The estimation of Arterio-Venous ratio (AVR) is an important phase in diagnosing various vascular diseases e.g. Diabetic Retinopathy. For calculating this value, it is essential to differentiate the vessels into arteries and veins. This paper presents a novel structural and automated method for artery/vein vessels classification in retinal images. Our method is tested on DRIVE database and the classification accuracy is 88.7 % for pixels and 89.07 % for vessel lines, respectively, which demonstrate the effectives of our approach. Our method will help to achieve the fundus disease surveillance on mobile and remote medical treatment. It has a remarkable social significance.
基金The authors would like to thank those who provided materials that were used in this study. This work was supported in part by the Natural Science Foundation of China under Grant 61472102, in part by the Fundamental Research Funds for the Central Universities under Grant HIT.NSRIF.2013091, and in part by the Humanity and Social Science Youth foundation of Ministry of Education of China under Grant 14YJC760001.
文摘Reliable detection of fundus lesion is important for automated screening of diabetic retinopathy. This paper presents a novel method to detect the fundus lesion in retinal fundus image based on a visual attention model. The proposed method intends to model the visual attention mechanism of ophthalmologists during observing fundus images. That is, the abnormal structures, such as the dark and bright lesions in the image, usually attract the most attention of experts, however, the normal structures, such as optic disc and vessels, have been usually selectively ignored. To measure the visual attention for abnormal and normal areas, the incremental coding length is computed in local and global manner respectively. The final saliency map of fundus lesion is a fusion of attention maps computed for the abnormal and normal areas. Experimental results conducted on the publicly DiaRetDB1 dataset show that the proposed method achieved a sensitivity of 0.71 at a specificity of 0.82 and an AUC of 0.76 for fundus lesion detection, and achieved an accuracy of 100% for normal area (optic disc) detection. The proposed method can assist the ophthalmologists in the inspection of fundus lesion.
