Master's student in Computer Science at the Federal University of Pelotas (UFPEL). Graduated in Computer Science from UFPEL. From 2021 to 2022, he participated in scientific initiation programs funded by CNPq, working on the project of cyber-physical systems for pest management. His research has focused on the areas of Artificial Intelligence, especially using computer vision to solve health problems, and also in the area of wireless sensor networks applied to pest management.
DRAT: A semi-supervised tool for automatic annotation of lesions caused by diabetic retinopathy Marcelo Dias, Carlos Santos, Marilton Aguiar, Daniel Welfer, Alejandro Pereira, et al. ACM International Conference Proceeding Series, 2024 Context: Diabetes is a significant global public health concern, with a growing number of affected. Patients with diabetes experience a reduced quality of life, primarily due to complications such as diabetic retinopathy. This complication, affecting a substantial portion of individuals with diabetes, is one of the leading causes of vision loss in adults. However, vision loss can be prevented through early diagnosis. Problem: Developing computational models for diagnosis is challenging due to the lack of datasets with adequate annotations, which are expensive and time-consuming to create. Solution: We introduce the Diabetic Retinopathy Annotation Tool, enabling automated annotation of retinal lesions in fundus images, expediting the process and allowing expert corrections. IS theory: This article incorporates ideas from Soft Systems Theory. Method: This research can be classified as explanatory, as it aims to establish a comprehensive theory by analyzing the results of experiments. This article employed a case study methodology to thoroughly examine fundus lesions, aiding in the creation of a tool for annotating and identifying these lesions. After, conducted experimental analysis to quantitatively evaluate the deep neural network model’s ability to predict and automatically label retinal lesions. Results: The model achieved an mAP of 0.4390 on the validation dataset and 0.3002 on the test dataset from the DDR dataset. Additionally, the tool demonstrated promising results when applied to the IDRID dataset, compared to actual lesions. Contributions and Impact in the IS area: This work introduces a tool for the healthcare field, potentially aiding in diagnosing diabetic retinopathy. The study also presents image processing techniques and computational model training methods applicable to future healthcare-oriented research.
An Improved Approach for Semantic Segmentation of Fundus Lesions using R2U-Net Alejandro Pereira, Carlos Santos, Marilton Aguiar, Daniel Welfer, Marcelo Dias, et al. Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society EMBS, 2024 Diabetic Retinopathy (DR) is a microvascular complication related to diabetes that affects approximately 33% of individuals with this condition and, if not detected and treated early, can lead to irreversible vision loss. Fundus lesions such as Hard and Soft Exudates, Hemorrhages, and Microaneurysms typically identify DR. The development of computational methods to segment these lesions plays a fundamental role in the early diagnosis of the disease. This paper proposes a new approach that uses an R2U-Net combined with data augmentation techniques for segmenting fundus lesions. We trained, adjusted, and evaluated the proposed work in the DDR dataset, achieving an accuracy of 99.87% and a mean Intersection over Union (mIoU) equal 59.69%. Furthermore, we assessed it in the IDRiD dataset, achieving an mIoU of 49.92%. The results obtained in the experiments highlight the potential contribution of the model in the lesion annotations for creating new DR datasets, which is essential given the scarcity of annotations in publicly available datasets.
Improved Detection of Fundus Lesions Using YOLOR-CSP Architecture and Slicing Aided Hyper Inference Alejandro Pereira, Carlos Santos, Marilton Aguiar, Daniel Welfer, Marcelo Dias, et al. IEEE Latin America Transactions, 2023 Diabetes affects millions of people worldwide, and diabetic retinopathy is a complication of diabetes. Brazil is the sixth in the world in the incidence of diabetes and has the highest numbers in Latin America with 15.7 million adults affected by this condition. Typically diabetic retinopathy is identified by lesions such as hard and soft exudates, microaneurysms, and vitreous hemorrhages. Early diagnosis of these lesions is essential to prevent the progression of the disease to severe stages and the consequent loss of vision. As the disease diagnosis is based on image analysis, it is possible to use deep learning models to detect artifacts in the retina. This article proposes a new method that uses a YOLOR-CSP architecture combined with the Slicing Aided Hyper Inference framework to detect fundus lesions. The proposed method was trained, adjusted, and evaluated using the DDR dataset, obtaining a mAP equal to 38.08%. The proposed method achieved values of AP equal to 40.90%, 46.60%, 26.10%, and 38.70% for hard exudates, soft exudates, microaneurysms, and vitreous hemorrhages, respectively, surpassing similar studies found in the literature.
A New Approach for Fundus Lesions Instance Segmentation Based on Mask R-CNN X101-FPN Pre-Trained Architecture Carlos Santos, Marilton Aguiar, Daniel Welfer, Marcelo Dias, Alejandro Pereira, et al. IEEE Access, 2023 Diabetic retinopathy is one of the main causes of vision loss, and it can be identified through ophthalmological examinations that aim to locate the presence of retinal lesions such as Microaneurysms, Hemorrhages, Soft Exudates, and Hard Exudates. The development of computerized approaches to perform the instance segmentation of these lesions can help in the early diagnosis of the disease. However, the segmentation of instances of artifacts in the retina is a complex task due to factors such as object size and morphological characteristics. This article proposes a new approach based on a Mask Regions with Convolutional Neural Network features (Mask R-CNN) architecture to perform instance segmentation of lesions associated with diabetic retinopathy. The proposed approach was trained, adjusted, and tested using different public datasets of diabetic retinopathy, which were implemented with the Detectron2 libraries and OpenCV. The best result obtained by the proposed approach in the Dataset for Diabetic Retinopathy (DDR) was using Tilling and Adam optimizer, reaching a mean Average Precision (<inline-formula> <tex-math notation="LaTeX">$mAP$ </tex-math></inline-formula>) of 0.2903 in the detection of fundus lesions for the limit of Intersection Over Union (<inline-formula> <tex-math notation="LaTeX">$IoU$ </tex-math></inline-formula>) of 0.5 in the validation stage and an <inline-formula> <tex-math notation="LaTeX">$mAP$ </tex-math></inline-formula> of 0.1670 in the detection of fundus lesions to the limit of <inline-formula> <tex-math notation="LaTeX">$IoU$ </tex-math></inline-formula> of 0.5 in the test step. The results obtained in the experiments demonstrate that the proposed approach presented promising results in the instance segmentation of microaneurysms with an increase in precision, which in our case reaches approximately 16%.
Detection of retinal microlesions through YOLOR-CSP architecture and image slicing with the SAHI algorithm Alejandro Pereira, Carlos Santos, Marilton Aguiar, Daniel Welfer, Marcelo Dias, et al. Proceedings of the International Joint Conference on Neural Networks, 2023 Diabetic retinopathy affects millions of working-age people worldwide. Of the countries in Latin America, Brazil has the highest incidence of cases. Diabetic retinopathy is detected through images of the fundus that contain lesions such as hard exudates, soft exudates, microaneurysms, and hemorrhages. Early identification of these lesions prevents the progression of the disease, which leads to a decrease in visual capacity. In addition, the early identification of these lesions allows the screening of patients who need priority care. The detection of these lesions occurs through the processing and analysis of fundus images using deep learning models. In this work, we present a new method that uses the You Only Learn One Representation with Cross Stage Partial Network (YOLOR-CSP) architecture combined with the Slicing Aided Hyper Inference (SAHI) framework to detect lesions. The proposed method was trained, adjusted, and evaluated using the Dataset for Diabetic Retinopathy (DDR) and the Indian Diabetic Retinopathy Image Dataset (IDRiD). The proposed method obtained in the data set DDR mAP equal to 38.08%, in the validation set, and 22.25% in the test set with SGD optimizer. The presented results were superior in the detection of eye fundus lesions in comparison with similar works found in the state-of-the-art literature.
A Method Based on Deep Neural Network for Instance Segmentation of Retinal Lesions Caused by Diabetic Retinopathy Carlos Santos, Marilton Aguiar, Daniel Welfer, Marcelo Silva, Alejandro Pereira, et al. Proceedings 2022 International Conference on Computational Science and Computational Intelligence Csci 2022, 2022 Diabetic Retinopathy is one of the main causes of vision loss and can be identified through ophthalmological exams that aim to locate the presence of retinal lesions such as microaneurysms, hemorrhages, soft exudates, and hard exudates. The development of computerized methods to perform the instance segmentation of lesions may support in the early diagnosis of the disease. However, the instance segmentation of retinal artifacts is a complex task due to factors such as the size of objects and their morphological characteristics. This article proposes a method based on a Mask R-CNN neural network architecture to perform instance segmentation of lesions associated with diabetic retinopathy. The proposed method was trained, adjusted, and tested using the public DDR and IDRiD Diabetic Retinopathy datasets, and implemented with the Detectron2 and OpenCV libraries. The proposed method reached in the DDR dataset, using the SGD optimizer, the mAP of 0.2660 for the limit of I oU of 0.5 in the validation step. The results obtained in the experiments demonstrate that the proposed method showed promising results in the instance segmentation of fundus lesions.
