An Effective Framework for Automatic Segmentation of Hard Exudates in Fundus Images

Yang, Nan; Lu, Huchuan; Fang, Guoliang; Yang, Gang

doi:10.1142/s0218126612500752

Cited by 5 publications

(2 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Some authors reported ROC curves; in that case, we also reported a single (sensitivity, specificity) pair: the one closest to the (sensitivity = 1, specificity = 1) coordinate. Note that all competing solutions (Kauppi et al, 2007;Yang et al, 2013;Franklin and Rajan, 2014;Kumar et al, 2014;Bharali et al, 2015;Mane et al, 2015;Dai et al, 2016) are trained at the lesion level, while ours is trained (in Kaggle-train) at the image level.…”

Section: Image-and Pixel-level Performance Of Convnetsmentioning

confidence: 99%

Deep image mining for diabetic retinopathy screening

Quellec

Charrière

Boudi

et al. 2017

Medical Image Analysis

381

243

View full text Add to dashboard Cite

Deep learning is quickly becoming the leading methodology for medical image analysis. Given a large medical archive, where each image is associated with a diagnosis, efficient pathology detectors or classifiers can be trained with virtually no expert knowledge about the target pathologies. However, deep learning algorithms, including the popular ConvNets, are black boxes: little is known about the local patterns analyzed by ConvNets to make a decision at the image level. A solution is proposed in this paper to create heatmaps showing which pixels in images play a role in the image-level predictions. In other words, a ConvNet trained for image-level classification can be used to detect lesions as well. A generalization of the backpropagation method is proposed in order to train ConvNets that produce high-quality heatmaps. The proposed solution is applied to diabetic retinopathy (DR) screening in a dataset of almost 90,000 fundus photographs from the 2015 Kaggle Diabetic Retinopathy competition and a private dataset of almost 110,000 photographs (e-ophtha). For the task of detecting referable DR, very good detection performance was achieved: A=0.954 in Kaggle's dataset and A=0.949 in e-ophtha. Performance was also evaluated at the image level and at the lesion level in the DiaretDB1 dataset, where four types of lesions are manually segmented: microaneurysms, hemorrhages, exudates and cotton-wool spots. For the task of detecting images containing these four lesion types, the proposed detector, which was trained to detect referable DR, outperforms recent algorithms trained to detect those lesions specifically, with pixel-level supervision. At the lesion level, the proposed detector outperforms heatmap generation algorithms for ConvNets. This detector is part of the Messidor® system for mobile eye pathology screening. Because it does not rely on expert knowledge or manual segmentation for detecting relevant patterns, the proposed solution is a promising image mining tool, which has the potential to discover new biomarkers in images.

show abstract

Section: Image-and Pixel-level Performance Of Convnetsmentioning

confidence: 99%

Deep image mining for diabetic retinopathy screening

Quellec

Charrière

Boudi

et al. 2017

Medical Image Analysis

381

243

View full text Add to dashboard Cite

show abstract

“…A multi-directional Gaussian matched filter is designed to detect the vessel network. For classifying the HEs or non-HEs, a set of 44 A novel method presented by Nan yang et al [30] for automatic segmentation of hard exudates in fundus images. Soft segmentation is used to segment pixels in the region into foreground and background.…”

Section: Image Processing Techniques For Detecting Exudatesmentioning

confidence: 99%

Hard Limiting and Soft Computing Techniques for Detection of Exudates in Retinal Images: A Futuristic Review

Kumar

Jayaram

Karegowda

2016

BJAST

View full text Add to dashboard Cite

Retinopathy in general, diabetic retinopathy in particular is the major concerns of blindness and appearance of hard exudates is one of its early symptoms. Since almost two decades, researchers are striving to develop automated algorithms and systems either to detect or to grade the severity of hard exudates. In this direction, this paper is a sincere effort to present the research carried out in this field. This paper has three specific intents i) to update the state-of-the-art techniques by bringing to the fore research that is being carried out, ii) to encourage prospective researchers in this area to explore various computational methodologies for the future development of detection techniques, and iii) to provide insights into some unsolved recognition problems.

show abstract