Search

Due to the presence of high glucose levels, diabetes mellitus (DM) is a widespread disease that can damage blood vessels in the retina and lead to loss of the visual system. To combat this disease, called Diabetic Retinopathy (DR), retinography, using images of the fundus of the retina, is the most used method for the diagnosis of Diabetic Retinopathy. The Deep Learning (DL) area achieved high performance for the classification of retinal images and even achieved almost the same human performance in diagnostic tasks. However, the performance of DL architectures is highly dependent on the optimal configuration of the hyperparameters. In this article, we propose the use of Neuroevolutionary Algorithms to optimize the hyperparameters corresponding to the DL model for the diagnosis of DR. The results obtained prove that the proposed method outperforms the results obtained by the classical approach.

Due to the presence of high glucose levels, diabetes mellitus (DM) is a widespread disease that can damage blood vessels in the retina and lead to loss of the visual system. To combat this disease, called Diabetic Retinopathy (DR), retinography, using images of the fundus of the retina, is the most used method for the diagnosis of Diabetic Retinopathy. The Deep Learning (DL) area achieved high performance for the classification of retinal images and even achieved almost the same human performance in diagnostic tasks. However, the performance of DL architectures is highly dependent on the optimal configuration of the hyperparameters. In this article, we propose the use of Neuroevolutionary Algorithms to optimize the hyperparameters corresponding to the DL model for the diagnosis of DR. The results obtained prove that the proposed method outperforms the results obtained by the classical approach.

Diabetic retinopathy is a complication of a widespread eye disease named diabetes mellitus. Diabetes mellitus, due to the increased glucose levels, may damage the retina’s blood vessels and cause visual complications and eventually blindness. Therefore, early detection and adequate assessment of disease progression are crucial for adequate treatment. The most widely used method for diagnosing diabetic retinopathy is the analysis of retinal fundus images obtained by retinography. Deep Learning-based methods have shown promising results as a diagnostic tool for diabetic retinopathy, achieving, in some cases, performance close to the human inspection of images. However, the performance of these methods relies heavily on fine-tuning the algorithm hyperparameters and big data sets. In this work, we propose training a Deep Learning network with evolutionary algorithms to classify three stages of Diabetic Retinopathy: i) no sign of diabetic retinopathy, ii) Non-proliferative diabetic retinopathy, and iii) proliferative diabetic retinopathy. We propose a neuroevolution methodology for selecting the most efficient Deep Learning model. The results of the neuroevolution methodology were improved by including Simulated Annealing strategies, Population Reinitialization and ensembles. With high accuracy, sensitivity, specificity and kappa index rates, the experiments show that our neuroevolution methodology for selecting the Deep Learning model hyperparameters is a competitive alternative for training deep neural networks to classify three stages of diabetic retinopathy even with a small data set.

Ocular toxoplasmosis (OT) is commonly diagnosed through the analysis of fundus images of the eye by a specialist. Despite Deep Learning being widely used to process and recognize pathologies in medical images, the diagnosis of ocular toxoplasmosis(OT) has not yet received much attention. A predictive computational model is a valuable time-saving option if used as a support tool for the diagnosis of OT. It could also help diagnose atypical cases, being particularly useful for ophthalmologists who have less experience. In this work, we propose the use of a deep learning model to perform automatic diagnosis of ocular toxoplasmosis from images of the eye fundus. A pretrained residual neural network is fine-tuned on a dataset of samples collected at the medical center of Hospital de Clínicas in Asunción, Paraguay. With sensitivity and specificity rates equal to 94% and 93%,respectively, the results show that the proposed model is highly promising. In order to replicate the results and advance further in this area of research, an open data set of images of the eye fundus labeled by ophthalmologists is made available.

Retinal images are widely used for diagnosis and eye disease detection. However, due to the acquisition process, retinal images often have problems such as low contrast, blurry details or artifacts. These problems may severely affect the diagnosis. Therefore, it is very important to enhance the visual quality of such images. Contrast enhancement is a pre-processing applied to images to improve their visual quality. This technique betters the identification of retinal structures in degraded retinal images. In this work, a novel algorithm based on multi-scale mathematical morphology is presented. First, the original image is blurred using the Open-Close Close-Open (OCCO) filter to reduce any artifacts in the image. Next, multiple bright and dark features are extracted from the filtered image by the Top-Hat transform. Finally, the maximum bright values are added to the original image and the maximum dark values are subtracted from the original image, previously adjusted by a weight. The algorithm was tested on 397 retinal images from the public STARE database. The proposed algorithm was compared with state of the art algorithms and results show that the proposal is more efficient in improving contrast, maintaining similarity with the original image and introducing less distortion than the other algorithms. According to ophthalmologists, the algorithm, by improving retinal images, provides greater clarity in the blood vessels of the retina and would facilitate the identification of pathologies.