A compactness based saliency approach for leakages detection in fluorescein angiogram
Summary (2 min read)
1 Introduction
- Fluorescein angiography (FA) is derived by taking a series of digital photographs of the retina before and after the fluorescein reaches the retinal circulation.
- Three types of leakage can be observed on fluorescein angiogram (FA) of malaria retinopathy (MR): large focal, punctate focal, and vessel segment leakage [4], see Fig.
- The saliency computation obtains a sparse representation of images, and an image can be classified as normal or abnormal (having bright lesions) by the obtained saliency information.
- The limitation of these works are that the effectiveness and robustness is not convincing enough, such as the dataset used for validation in [12] is relative too small.
- The leaking regions may be defined as salient regions as the leakage of fluorescent dye causes a large difference in brightness between the leak and surrounding non-leaking areas.
2 Method
- The proposed framework consists of three main phases, which are saliency detection, saliency refinement, and leakage detection.
- This is an advantage of the relative distance metric in reflecting the relative density of points and relative scale of the imaged objects.
- Therefore, this section introduces an additional feature - compactness.
- The measure of compactness of an object might be used as a complementary feature for saliency measurement, with the aim of overcoming the conflict of the falsely detected salient region.
- Any small and/or isolated objects are eliminated after masking the vessel regions, by the use of a disk-shaped opening operation with a radius of 2 pixels.
3 Experimental Results
- The detection framework in three types of leakage was implemented in Matlab 2013a.
- The dataset the authors used comprises retinal FA images taken from children with CM admitted to the Malaria Research Project Ward, Queen Elizabeth Central Hospital, Blantyre, Malawi.
- Consent was given by the guardians of subjects before examination and imaging.
- 20 images (one per patient) with large focal leakage, and 10 images from 6 patients with punctate leakage were used.
- The sensitivity, false positive per image, false negative per image, and ratio of overlapping area (OR) were used as the evaluation metrics.
3.2.1 Punctate focal
- The top row of Fig. 4 show that the punctate focal leak regions were detected by both the human expert and the proposed method, respectively.
- Table 1 illustrates the evaluation results of their automated method on detecting punctate focal leakage, in terms of sensitivity, false positives (FP) per image of punctate focal leakage number, and the false negatives (FN) per image of punctate focal leakage number.
- According to the human reference standard there were 240 sites of punctate focal leakage in the tested 10 images.
- 29 site were missed, leading to a false negative ratio of 2.832 per image.
- Meanwhile, false positive ratio is 1.1 per image, because 11 sites were falsely determined as punctate focal leakage by proposed method due to the image artefact and local imbalanced illumination.
3.2.2 Large focal
- The middle row of Fig. 4 demonstrates the detection of large focal leakage by both human expert and the proposed method, respectively.
- It can be seen that the detected leakage regions are slightly different, since it is difficult to exactly define the boundary of a leaking area by hand since the contrast gradually fades at the edge of the lesion.
- Table 1 also indicates the evaluation results of the developed framework in the detection of large focal leakage, in terms of sensitivity, false positives of focal leakage regions per image, false negatives per image of focal leakage regions, and the overlapping area between the regions detected by the proposed method and human expert.
- In these studied images, a total of 41 sites of large focal leak were identified by the human expert reference standard.
- In their framework, it only failed to detect 1 out of these focal leak sites, which has been evidenced by the sensitivity ratio of 0.976, and has a false negative ratio of 0.05 per image.
3.2.3 Vessel segment
- The bottom of Fig. 4 indicate the results of vessel segment leakage detection.
- An AUC of 1.0 means that the classifier distinguishes class examples perfectly.
- In order to assess inter-observer variation, a third grader labeled the vessels using the same method.
- In brief, this table shows that the developed automatic vessel segments leakage detection method can perform better than or at least as well as a human expert.
4 Conclusions
- This paper proposed an innovative framework for the detection of three types of leakage founded on saliency detection, with the aim of supporting the study of abnormalities revealed in retinal images.
- The framework benefits from three major components: saliency estimation, saliency refinement, and leakage detection.
- The authors have successfully integrated compactness based feature to refined the intensity based saliency map.
- The effectiveness of this method has been tested on FA images from patients with malarial retinopathy.
- The authors method demonstrated satisfactory overall performance on large focal, punctate focal and vessel segment leakage detection.
Did you find this useful? Give us your feedback
Citations
7 citations
Cites methods from "A compactness based saliency approa..."
...Most of the commonly used methods (Zhao et al., 2017, 2015) to generate saliency map of FFA image are pixel-based methods, which are time-consuming and thus do not meet the requirements of our work....
[...]
7 citations
1 citations
References
86 citations
84 citations
65 citations
62 citations
"A compactness based saliency approa..." refers background in this paper
...In other ocular diseases has received little, such as diabetic retinopathy [6], retinal vein occlusion [7], hyperfluorescent [8] and choroidal neovascularization [9–11], if any, attention....
[...]
48 citations
Related Papers (5)
Frequently Asked Questions (2)
Q2. What are the future works mentioned in the paper "A compactness based saliency approach for leakages detection in fluorescein angiogram" ?
In their future work, the authors will extend the current saliency-guided leakage detection model to other vascular diseases, such as diabetic retinopathy. The authors believe that this innovative framework has the potential to be developed further as a useful tool for fast, accurate and objective assessment of leak in a range of vascular diseases.