arXiv subjects

Cesar Henrique Comin

Publications and source records attributed to Cesar Henrique Comin.

8 recordsLinked to original sources

arXiv2026

Evaluation of Blood Vessel Segmentation Methods on Hard-to-Detect Vascular Structures

Due to the intricate structure of vascular trees, minor segmentation errors can significantly alter connectivity patterns and increase variability in extracted morphological properties. Global metrics such as the Dice coefficient, precision, and recall often overlook inaccuracies in specific regions of a sample. To address this, we define a Local Vessel Salience (LVS) index to quantify the difficulty of identifying specific vessel segments. This index is used to evaluate the performance of 16 segmentation methods across six widely used 2D datasets. The LVS index is calculated for each vessel pixel by comparing local vessel intensity against the surrounding background. We introduce a metric termed mean Low-Salience Recall (mLSR) to quantify how effectively algorithms recover hard-to-detect vessels. Furthermore, we propose a proof-of-concept data augmentation procedure guided by the LVS index aimed at improving neural network segmentation performance. Our findings demonstrate that segmentation performance strongly correlates with LVS, revealing systematic errors in vessels with low salience. The mLSR across all evaluated methods was significantly lower than standard recall values, with a typical performance decrease of 20 percentage points. In benchmark datasets such as DRIVE and OCTA-500, mLSR values for the top-performing methods were approximately 58%. The developed methodology provides a quantitative basis for the design of segmentation algorithms with improved sensitivity to hard-to-detect vessels and enhanced capabilities for preserving vascular connectivity.

cs.CV↗

arXiv2026

Investigating the Visual Cues of CNNs for Vascular Segmentation: A Case Study in Microscopy and Fundus Imaging

Vascular segmentation is a standard procedure for clinical diagnosis, yet the specific visual features determining model decisions remain poorly understood. This paper investigates the visual cues Convolutional Neural Networks (CNNs) use to segment blood vessels across two distinct imaging domains: fluorescence microscopy and retinal fundus photography. We employ a series of experiments to quantify the influence of shape, texture, and receptive field on segmentation performance. First, we isolate texture and intensity by evaluating performance on patches subjected to pixel shuffling and normalization. Second, we assess global shape relevance by training models on sparse contours and centerlines. Lastly, we quantify the required spatial context by systematically varying the network's theoretical and effective receptive fields. Within the scope of the evaluated datasets, we found that pixel intensity is more relevant than texture, though networks maintain surprisingly high accuracy even when both cues are removed. Furthermore, CNNs struggle to extrapolate full vessel geometry from shape cues alone, typically relying on a relatively small effective receptive field of around 20 pixels, though global context provides a modest benefit for fundus images. While specific to the modalities studied, this methodology offers a quantitative foundation to audit and refine deep learning systems in vascular imaging.

eess.IV↗

arXiv2024

A new dataset for measuring the performance of blood vessel segmentation methods under distribution shifts

Creating a dataset for training supervised machine learning algorithms can be a demanding task. This is especially true for medical image segmentation since one or more specialists are usually required for image annotation, and creating ground truth labels for just a single image can take up to several hours. In addition, it is paramount that the annotated samples represent well the different conditions that might affect the imaged tissues as well as possible changes in the image acquisition process. This can only be achieved by considering samples that are typical in the dataset as well as atypical, or even outlier, samples. We introduce VessMAP, a heterogeneous blood vessel segmentation dataset acquired by carefully sampling relevant images from a larger non-annotated dataset. A methodology was developed to select both prototypical and atypical samples from the base dataset, thus defining an assorted set of images that can be used for measuring the performance of segmentation algorithms on samples that are highly distinct from each other. To demonstrate the potential of the new dataset, we show that the validation performance of a neural network changes significantly depending on the splits used for training the network.

cs.CV↗

arXiv2020

Contrarian effects and echo chamber formation in opinion dynamics

The relationship between the topology of a network and specific types of dynamics unfolding in networks constitutes a subject of substantial interest. One type of dynamics that has attracted increasing attention because of its several potential implications is opinion formation. A phenomenon of particular importance, known to take place in opinion formation, is echo chambers' appearance. In the present work, we approach this phenomenon, while emphasizing the influence of contrarian opinions in a multi-opinion scenario. To define the contrarian opinion, we considered the Underdog effect, which is the eventual tendency of people to support the less popular option. We also considered an adaptation of the Sznajd dynamics with the possibility of friendship rewiring, performed on several network models. We analyze the relationship between topology and opinion dynamics by considering two measurements: opinion diversity and network modularity. Two specific situations have been addressed: (i) the agents can reconnect only with others sharing the same opinion; and (ii) same as in the previous case, but with the agents reconnecting only within a limited neighborhood. This choice can be justified because, in general, friendship is a transitive property along with subsequent neighborhoods (e.g., two friends of a person tend to know each other). As the main results, we found that the Underdog effect, if strong enough, can balance the agents' opinions. On the other hand, this effect decreases the possibilities of echo-chamber formation. We also found that the restricted reconnection case reduced the chances of echo chamber formation and led to smaller echo chambers.

physics.soc-ph↗

arXiv2019

Opinion Diversity and Social Bubbles in Adaptive Sznajd Networks

Among the several approaches that have been attempted at studying opinion dynamics, the Sznajd model provides some particularly interesting features, such as its simplicity and ability to represent some of the mechanisms believed to be involved in opinion dynamics. The standard Sznajd model at zero temperature is characterized by converging to one stable state, implying null diversity of opinions. In the present work, we develop an approach -- namely the adaptive Sznajd model -- in which changes of opinion by an individual (i.e. a network node) implies in possible alterations in the network topology. This is accomplished by allowing agents to change their connections preferentially to other neighbors with the same state. The diversity of opinions along time is quantified in terms of the exponential of the entropy of the opinions density. Several interesting results are reported, including the possible formation of echo chambers or social bubbles. Additionally, depending on the parameters configuration, the dynamics may converge to different equilibrium states for the same parameter setting, which suggests that this phenomenon can be a phase transition. The average degree of the network strongly influences the resultant opinion distribution, which means that echo chambers are easily formed in lower connected systems.

physics.soc-ph↗

arXiv2016

A diffusion-based approach to obtaining the borders of urban areas

The access to an ever increasing amount of information in the modern world gave rise to the development of many quantitative indicators about urban regions in the globe. Therefore, there is a growing need for a precise definition of how to delimit urban regions, so as to allow proper respective characterization and modeling. Here we present a straightforward methodology to automatically detect urban region borders around a single seed point. The method is based on a diffusion process having street crossings and terminations as source points. We exemplify the potential of the methodology by characterizing the geometry and topology of 21 urban regions obtained from 8 distinct countries. The geometry is studied by employing the lacunarity measurement, which is associated to the regularity of holes contained in a pattern. The topology is analyzed by associating the betweenness centrality of the streets with their respective class, such as motorway or residential, obtained from a database.

physics.soc-ph↗

arXiv2016

Identifying the starting point of a spreading process in complex networks

When dealing with the dissemination of epidemics, one important question that can be asked is the location where the contamination began. In this paper, we analyze three spreading schemes and propose and validate an effective methodology for the identification of the source nodes. The method is based on the calculation of the centrality of the nodes on the sampled network, expressed here by degree, betweenness, closeness and eigenvector centrality. We show that the source node tends to have the highest measurement values. The potential of the methodology is illustrated with respect to three theoretical complex network models as well as a real-world network, the email network of the University Rovira i Virgili.

physics.soc-ph↗

arXiv2015

Minimal paths between communities induced by geographical networks

In this work we investigate the betweenness centrality in geographical networks and its relationship with network communities. We show that vertices with large betweenness define what we call characteristic betweenness paths in both modeled and real-world geographical networks. We define a geographical network model that possess a simple topology while still being able to present such betweenness paths. Using this model, we show that such paths represent pathways between entry and exit points of highly connected regions, or communities, of geographical networks. By defining a new network, containing information about community adjacencies in the original network, we describe a means to characterize the mesoscale connectivity provided by such characteristic betweenness paths.

physics.soc-ph↗