Dora Elisa Alvarado-Carrillo

Blood vessel segmentation in X-ray coronary angiography (XCA) plays a crucial role in diagnosing cardiovascular diseases, enabling a precise assessment of arterial structures. However, segmentation is challenging due to a low signal-to-noise ratio, interfering background structures, and vessel bifurcations, which hinder the accuracy of deep learning models. Additionally, deep learning models for this task often require high computational resources, limiting their practical application in real-time clinical settings. This study proposes a lightweight variant of the U-Net architecture using a structured kernel pruning strategy inspired by the Lottery Ticket Hypothesis. The pruning method systematically removes entire convolutional filters from each layer based on a global reduction factor, generating compact subnetworks that retain key representational capacity. This results in a significantly smaller model without compromising the segmentation performance. This approach is evaluated on two benchmark datasets, demonstrating consistent improvements in segmentation accuracy compared to the vanilla U-Net. Additionally, model complexity is significantly reduced from 31 M to 1.9 M parameters, improving efficiency while maintaining high segmentation quality.

Coronary stenosis results from unnatural narrowing of the heart arteries due to the accumulation of adipose depots, leading to different heart diseases and yielding top mortality worldwide. Thus far, deep learning-based methods for automatic stenosis over X-ray Coronary Angiography (XCA) have employed state-of-the-art architectures to solve the ImageNet challenge. With the advance of deep learning, contemporary architectures incorporated a variety of attention mechanisms to improve performance. Therefore, this paper presents a study of three attention mechanisms for stenosis detection in XCA images. Extensive experiments and comparisons over different Residual backbone networks are presented to verify the effectiveness of including such attention modules. An improvement of 4%, 10%, and 10% on the accuracy, recall, and F1-score was achieved using the approach, reaching mean values of 0.8787, 0.8610, and 0.8732, respectively.

Introducción: La alta prevalencia de Diabetes Mellitus tipo 2 en México ha posicionado a la retinopatía diabética como la principal causa de ceguera en adultos en edad productiva en México. Por ello, la detección oportuna de este padecimiento es una tarea prioritaria para el sistema público de salud. En el presente artículo se estudia el desempeño de un nuevo algoritmo para la determinación de la forma de la arcada temporal mayor de la retina, mediante el uso de técnicas de segmentación de imágenes y modelado numérico de curvas. Método: La metodología propuesta emplea Filtros Gaussianos de Correspondencia que realzan la geometría de los vasos sanguíneos. Posteriormente, la estructura vascular es segmentada mediante la umbralización global de la imagen realzada. Dicha segmentación es utilizada como entrada para construir un modelo numérico de las arcadas temporales superior en inferior, utilizando funciones Spline. Resultados: La evaluación de desempeño se realizó utilizando 136 imágenes de pixeles. El algoritmo de segmentación automática de venas de la retina mediante el método GMF obtuvo un valor de Accuracy de 0.9852; el algoritmo de modelado numérico dio un resultado de 6.01 pixeles en la métrica de la distancia media al punto más cercano (MDCP). Otro estudio previo reportó 12.33 pixeles. Con respecto al tiempo, se reportó un tiempo promedio de 10.65 segundos por imagen. Discusión: El método propuesto fue capaz de realizar eficientemente el modelado numérico de las arcadas temporales en imágenes de fondo de ojo. Los resultados demuestran que este método es una herramienta computacional útil para el diagnóstico de alteraciones en la anatomía del ojo.

Automating retinal vessel segmentation is a primary element of computer-aided diagnostic systems for many retinal diseases. It facilitates the inspection of shape, width, tortuosity, and other blood vessel characteristics. In this paper, a new method that incorporates Distorted Gaussian Matched Filters (D-GMFs) with adaptive parameters as part of a Deep Convolutional Architecture is proposed. The D-GaussianNet includes D-GMF units, a variant of the Gaussian Matched Filter that considers curvature, placed at the beginning and end of the network to implicitly indicate that spatial attention should focus on curvilinear structures in the image. Experimental results on datasets DRIVE, STARE, and CHASE show state-of-the-art performance with an accuracy of 0.9565, 0.9647, and 0.9609 and a F1-score of 0.8233, 0.8141, and 0.8077, respectively.