Automatic segmentation of pigment deposits in retinal fundus images of Retinitis Pigmentosa
Introduction
Many eye diseases, such as retinal degeneration, optic atrophy and eye malformations, are caused by genetic eye disorders. In particular, retinitis pigmentosa (RP) comprises a group of inherited retinal diseases that causes abnormalities of photoreceptors rods and cones) and retinal pigment ephitelium (Hartong et al., 2006). The prevalence is about 1 in 4000 and worldwide incidence of about 1 million individuals. Rods are the predominantly affected photoreceptors and dysfunction causes night blindness and peripheral field loss beginning as early as the teenage years. Progression leads to central acuity loss and legal blindness in the majority of patients. In typical form patients can have a normal fundus appearance in the early stages. In more advanced stages, signs include attenuated retinal vessels, pigment deposits, and pallor of the optic discs. The pigment deposits (PDs) are distributed circumferentially around the mid periphery in the zone where rods normally are at maximum concentration. Diagnosis of the RP is made by ophthalmologists starting from clinical observation and is confirmed by some tests: (i) the fundus camera, that allows a photographic record of the retinal images, useful for the diagnosis and follow-up of the disease; (ii) the electroretinogram (ERG), which measures the electrical activity of the retina; (iii) and Optical Coherence Tomography (OCT), that allows the evaluation the thickness of the macular region, preservation or the alteration of the different retinal layers. Further confirmations can be genetic analysis. Actually, there is no therapy that can be used to stop the degeneration of the RP, but if an early diagnosis is performed, therapeutic strategies aimed at slowing down the degeneration process can be adopted. Recently, due to the improvement in computing technology, an increasing interest has been focused on automatic systems supporting medical diagnosis. For retinal diseases, automatic methods for image analysis have been developed to support ophthalmologists in the diagnosis, most of which being based on the analysis of fundus images (Teng et al., 2002, Das et al., 2014, Rahebi and Hardalaç, 2016, Sowmya, 2016). Some methods allow a diagnosis of a disease, such as diabetic retinopathy (Akram and Khan, 2012, Ganesan et al., 2014) or age-related macular degeneration (Mookiah et al., 2014, Mittal and Kumari, 2015). The current literature offers some methods based on the analysis of OCT to quantify RP and to track its progression (Hood et al., 2011, Yang et al., 2011, Menghini et al., 2015). To the best of our knowledge, the literature about the automatic detection of PDs simply by analyzing fundus images is extremely limited (Das et al., 2014), whereas the diagnosis by fundus camera represents a more convenient, less invasive, easily repeatable and faster technique, with no contraindications. In particular, there are contexts where low-cost and easy-access diagnostic systems play a crucial role, such as in resource-limited regions and countries. A further reason for choosing to use fundus images consists in the advent of low-cost handheld fundus cameras (the HF-camera) as well as external optics that turn a smartphone into a fundus camera, so providing a valid alternative option for retinal image acquisition in resource-limited settings. Retinitis pigmentosa in fundus images presents as an attenuation of the vessels, a waxy disc pallor and the presence of pigment deposits. Even if a number of approaches have been proposed to analyze the retinal vessel structure (Frucci et al., 2016, Goswami et al., 2017) and the pallor of the optic disk (Nakano et al., 2016), virtually no methods addressing the latter symptom, namely pigment deposits, can be found in literature. In this paper, a PD automatic detection method based on the analysis of the fundus image is proposed. By applying a segmentation technique based on watershed transformation, the fundus image is partitioned into regions, which undergo a classification process to discriminate regions belonging to PDs from regions with a normal fundus. The classification is mainly based on the analysis of the features associated with the regions and their relationships. The approach has been tested on a property dataset of fundus images, which are provided with two ground truths manually segmented by two different specialists.
The rest of the paper is organized as follows: in Section 2, the proposed method is described, after some preliminary considerations; Section 3 contains details relating to the results obtained; Section 4, contains a discussion about obtained results; and finally in Section 5, some conclusions are drawn.
Section snippets
Method
The automatic segmentation of PDs is a challenging task due to several different factors, including both the acquisition protocol and the peculiar structures of the retinal fundus. The retina is a spherical surface that is imaged by an acquisition device projecting a single illuminating source. Hence, the different incidence angle of the light between the central and peripheral regions of the fundus introduces color and resolution degradation when moving from the center towards the periphery.
Results
The experiments have been performed on twenty images of subjects affected by RP. Two specialists with different levels of confidence have manually annotated each image, so providing the corresponding ground truths (GT1 and GT2) for a qualitative and quantitative evaluation of the proposed method. A qualitative assessment of the proposed method is shown in Figs. 9 and 10 for eight input images.
From a quantitative point of view, the performances have been measured according to standard indexes,
Discussion
Fundus images are characterized by a great number of distortions, which makes segmentation and analysis tasks particularly challenging. Retinal vessel segmentation and optical disk detection are just two examples of particularly complex tasks that have already been widely investigated in the literature. In the specific case of Retinitis Pigmentosa, the detection of pigment deposits is even more challenging, as PDs exhibit a very high degree of variability in color and shape. For these reasons,
Conclusions
In this paper, we have addressed the problem of the automatic segmentation of pigment deposits in fundus images of patients suffering from RP. This disease usually causes vision loss and may lead to complete blindness. Although at present no cure exists, its progression can be slowed down by therapeutic strategies if it is diagnosed before it reaches an advanced stage. The main contribution of this work is to provide a segmentation technique for the detection of PDs directly in fundus retinal
Conflict of interest statement
No conflict of interest.
References (27)
- et al.
Size invariant circle detection
Image Vis. Comput.
(1999) - et al.
Severe: segmenting vessels in retina images
Pattern Recogn. Lett.
(2016) - et al.
Retinitis pigmentosa
Lancet
(2006) Topographic distance and watershed lines
Signal Process.
(1994)- et al.
Automated detection and segmentation of drusen in retinal fundus images
Comput. Electr. Eng.
(2015) - et al.
Exudate detection in color retinal images for mass screening of diabetic retinopathy
Med. Image Anal.
(2014) Contrast limited adaptive histogram equalization
- et al.
Automated detection of dark and bright lesions in retinal images for early detection of diabetic retinopathy
J. Med. Syst.
(2012) - et al.
An expert system to distinguish a defective eye from a normal eye
2014 International Conference on Issues and Challenges in Intelligent Computing Techniques (ICICT)
(2014) Oversegmentation reduction by flooding regions and digging watershed lines
Int. J. Pattern Recogn. Artif. Intell.
(2006)
Computer-aided diabetic retinopathy detection using trace transforms on digital fundus images
Med. Biol. Eng. Comput.
Automatic retina exudates segmentation without a manually labelled training set
2011 IEEE International Symposium on Biomedical Imaging: From Nano to Macro
Automatic measurement and analysis of vessel width in retinal fundus image
2017 First International Conference on Intelligent Computing and Communication
Cited by (9)
A deep convolutional neural network for diabetic retinopathy detection via mining local and long-range dependence
2024, CAAI Transactions on Intelligence TechnologyX-Linked Retinitis Pigmentosa Gene Therapy: Preclinical Aspects
2023, Ophthalmology and TherapyUAU-Net: United Attention U-Shaped Network for the Segmentation of Pigment Deposits in Fundus Images of Retinitis Pigmentosa
2023, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)Deep Learning-Based Pipeline for the Detection of Multiple Ocular Diseases
2023, Lecture Notes in Electrical EngineeringConstruction and verification of color fundus image retinal vessels segmentation algorithm under BP neural network
2021, Journal of Supercomputing