RESUMO

Glaucoma is an eye disease that gradually deteriorates vision. Much research focuses on extracting information from the optic disc and optic cup, the structure used for measuring the cup-to-disc ratio. These structures are commonly segmented with deeplearning techniques, primarily using Encoder-Decoder models, which are hard to train and time-consuming. Object detection models using convolutional neural networks can extract features from fundus retinal images with good precision. However, the superiority of one model over another for a specific task is still being determined. The main goal of our approach is to compare object detection model performance to automate segment cups and discs on fundus images. This study brings the novelty of seeing the behavior of different object detection models in the detection and segmentation of the disc and the optical cup (Mask R-CNN, MS R-CNN, CARAFE, Cascade Mask R-CNN, GCNet, SOLO, Point_Rend), evaluated on Retinal Fundus Images for Glaucoma Analysis (REFUGE), and G1020 datasets. Reported metrics were Average Precision (AP), F1-score, IoU, and AUCPR. Several models achieved the highest AP with a perfect 1.000 when the threshold for IoU was set up at 0.50 on REFUGE, and the lowest was Cascade Mask R-CNN with an AP of 0.997. On the G1020 dataset, the best model was Point_Rend with an AP of 0.956, and the worst was SOLO with 0.906. It was concluded that the methods reviewed achieved excellent performance with high precision and recall values, showing efficiency and effectiveness. The problem of how many images are needed was addressed with an initial value of 100, with excellent results. Data augmentation, multi-scale handling, and anchor box size brought improvements. The capability to translate knowledge from one database to another shows promising results too.

RESUMO

ABSTRACT Segmentation is vital in Optical Coherence Tomography Angiography (OCT-A) images. The separation and distinction of the different parts that build the macula simplify the subsequent detection of observable patterns/illnesses in the retina. In this work, we carried out multi-class image segmentation where the best characteristics are highlighted in the appropriate plexuses by comparing different neural network architectures, including U-Net, ResU-Net, and FCN. We focus on two critical zones: retinal vasculature (RV) and foveal avascular zone (FAZ). The precision obtained from the RV and FAZ segmentation over 316 OCT-A images from the OCT-A 500 database at 93.21% and 92.59%, where the FAZ was segmented with an accuracy of 99.83% for binary classification.

RESUMEN La segmentación juega un papel vital en las imágenes de angiografía por tomografía de coherencia óptica (OCT-A), ya que la separación y distinción de las diferentes partes que forman la mácula simplifican la detección posterior de patrones/enfermedades observables en la retina. En este trabajo, llevamos a cabo una segmentación de imágenes multiclase donde se destacan las mejores características en los plexos apropiados al comparar diferentes arquitecturas de redes neuronales, incluidas U-Net, ResU-Net y FCN. Nos centramos en dos zonas críticas: la segmentación de la vasculatura retiniana (RV) y la zona avascular foveal (FAZ). La precisión para RV y FAZ en 316 imágenes OCT-A de la base de datos OCT-A 500 se obtuvo en 93.21 % y 92.59 %. Cuando se segmentó la FAZ en una clasificación binaria, con un 99.83% de precisión.

RESUMO

ABSTRACT Deep learning (DL) techniques achieve high performance in the detection of illnesses in retina images, but the majority of models are trained with different databases for solving one specific task. Consequently, there are currently no solutions that can be used for the detection/segmentation of a variety of illnesses in the retina in a single model. This research uses Transfer Learning (TL) to take advantage of previous knowledge generated during model training of illness detection to segment lesions with encoder-decoder Convolutional Neural Networks (CNN), where the encoders are classical models like VGG-16 and ResNet50 or variants with attention modules. This shows that it is possible to use a general methodology using a single fundus image database for the detection/segmentation of a variety of retinal diseases achieving state-of-the-art results. This model could be in practice more valuable since it can be trained with a more realistic database containing a broad spectrum of diseases to detect/segment illnesses without sacrificing performance. TL can help achieve fast convergence if the samples in the main task (Classification) and sub-tasks (Segmentation) are similar. If this requirement is not fulfilled, the parameters start from scratch.

RESUMEN Las técnicas de Deep Learning (DL) han demostrado un buen desempeño en la detección de anomalías en imágenes de retina, pero la mayoría de los modelos son entrenados en diferentes bases de datos para resolver una tarea en específico. Como consecuencia, actualmente no se cuenta con modelos que se puedan usar para la detección/segmentación de varias lesiones o anomalías con un solo modelo. En este artículo, se utiliza Transfer Learning (TL) con la cual se aprovecha el conocimiento adquirido para determinar si una imagen de retina tiene o no una lesión. Con este conocimiento se segmenta la imagen utilizando una red neuronal convolucional (CNN), donde los encoders o extractores de características son modelos clásicos como VGG-16 y ResNet50 o variantes con módulos de atención. Se demuestra así, que es posible utilizar una metodología general con bases de datos de retina para la detección/ segmentación de lesiones en la retina alcanzando resultados como los que se muestran en el estado del arte. Este modelo puede ser entrenado con bases de datos más reales que contengan una gama de enfermedades para detectar/ segmentar sin sacrificar rendimiento. TL puede ayudar a conseguir una convergencia rápida del modelo si la base de datos principal (Clasificación) se parece a la base de datos de las tareas secundarias (Segmentación), si esto no se cumple los parámetros básicamente comienzan a ajustarse desde cero.

RESUMO

Segmenting vessels in brain images is a critical step for many medical interventions and diagnoses of illnesses. Recent advances in artificial intelligence provide better models, achieving a human-like level of expertise in many tasks. In this paper, we present a new approach to segment Time-of-Flight Magnetic Resonance Angiography (TOF-MRA) images, relying on fewer training samples than state-of-the-art methods. We propose a conditional generative adversarial network with an adapted generator based on a concatenated U-Net with a residual U-Net architecture (UUr-cGAN) to carry out blood vessel segmentation in TOF-MRA images, relying on data augmentation to diminish the drawback of having few volumes at disposal for training the model, while preventing overfitting by using regularization techniques. The proposed model achieves 89.52% precision and 87.23% in Dice score on average from the cross-validated experiment for brain blood vessel segmentation tasks, which is similar to other state-of-the-art methods while using considerably fewer training samples. UUr-cGAN extracts important features from small datasets while preventing overfitting compared to other CNN-based methods and still achieve a relatively good performance in image segmentation tasks such as brain blood vessels from TOF-MRA.

RESUMO

Dementia is a neurodegenerative disease that leads to the development of cognitive deficits, such as aphasia, apraxia, and agnosia. It is currently considered one of the most significant major medical problems worldwide, primarily affecting the elderly. This condition gradually impairs the patient's cognition, eventually leading to the inability to perform everyday tasks without assistance. Since dementia is an incurable disease, early detection plays an important role in delaying its progression. Because of this, tools and methods have been developed to help accurately diagnose patients in their early stages. State-of-the-art methods have shown that the use of syntactic-type linguistic features provides a sensitive and noninvasive tool for detecting dementia in its early stages. However, these methods lack relevant semantic information. In this work, we propose a novel methodology, based on the semantic features approach, by using sentence embeddings computed by Siamese BERT networks (SBERT), along with support vector machine (SVM), K-nearest neighbors (KNN), random forest, and an artificial neural network (ANN) as classifiers. Our methodology extracted 17 features that provide demographic, lexical, syntactic, and semantic information from 550 oral production samples of elderly controls and people with Alzheimer's disease, provided by the DementiaBank Pitt Corpus database. To quantify the relevance of the extracted features for the dementia classification task, we calculated the mutual information score, which demonstrates a dependence between our features and the MMSE score. The experimental classification performance metrics, such as the accuracy, precision, recall, and F1 score (77, 80, 80, and 80%, respectively), validate that our methodology performs better than syntax-based methods and the BERT approach when only the linguistic features are used.