César H. Comin

Matheus Viana da Silva, Natália de Carvalho Santos, Julie Ouellette et al.

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.

Cesar H. Comin, Wesley N. Galvão

Semantic segmentation of blood vessels is an important task in medical image analysis, but its progress is often hindered by the scarcity of large annotated datasets and the poor generalization of models across different imaging modalities. A key aspect is the tendency of Convolutional Neural Networks (CNNs) to learn texture-based features, which limits their performance when applied to new domains with different visual characteristics. We hypothesize that leveraging geometric priors of vessel shapes, such as their tubular and branching nature, can lead to more robust and data-efficient models. To investigate this, we introduce VessShape, a methodology for generating large-scale 2D synthetic datasets designed to instill a shape bias in segmentation models. VessShape images contain procedurally generated tubular geometries combined with a wide variety of foreground and background textures, encouraging models to learn shape cues rather than textures. We demonstrate that a model pre-trained on VessShape images achieves strong few-shot segmentation performance on two real-world datasets from different domains, requiring only four to ten samples for fine-tuning. Furthermore, the model exhibits notable zero-shot capabilities, effectively segmenting vessels in unseen domains without any target-specific training. Our results indicate that pre-training with a strong shape bias can be an effective strategy to overcome data scarcity and improve model generalization in blood vessel segmentation.

João Pedro Parella, Matheus Viana da Silva, Cesar Henrique Comin

Many studies regarding the vasculature of biological tissues involve the segmentation of the blood vessels in a sample followed by the creation of a graph structure to model the vasculature. The graph is then used to extract relevant vascular properties. Small segmentation errors can lead to largely distinct connectivity patterns and a high degree of variability of the extracted properties. Nevertheless, global metrics such as Dice, precision, and recall are commonly applied for measuring the performance of blood vessel segmentation algorithms. These metrics might conceal important information about the accuracy at specific regions of a sample. To tackle this issue, we propose a local vessel salience (LVS) index to quantify the expected difficulty in segmenting specific blood vessel segments. The LVS index is calculated for each vessel pixel by comparing the local intensity of the vessel with the image background around the pixel. The index is then used for defining a new accuracy metric called low-salience recall (LSRecall), which quantifies the performance of segmentation algorithms on blood vessel segments having low salience. The perspective provided by the LVS index is used to define a data augmentation procedure that can be used to improve the segmentation performance of convolutional neural networks. We show that segmentation algorithms having high Dice and recall values can display very low LSRecall values, which reveals systematic errors of these algorithms for vessels having low salience. The proposed data augmentation procedure is able to improve the LSRecall of some samples by as much as 25%. The developed methodology opens up new possibilities for comparing the performance of segmentation algorithms regarding hard-to-detect blood vessels as well as their capabilities for vascular topology preservation.

Matheus V. da Silva, Julie Ouellette, Baptiste Lacoste et al.

Characterizing blood vessels in digital images is important for the diagnosis of many types of diseases as well as for assisting current researches regarding vascular systems. The automated analysis of blood vessels typically requires the identification, or segmentation, of the blood vessels in an image or a set of images, which is usually a challenging task. Convolutional Neural Networks (CNNs) have been shown to provide excellent results regarding the segmentation of blood vessels. One important aspect of CNNs is that they can be trained on large amounts of data and then be made available, for instance, in image processing software for wide use. The pre-trained CNNs can then be easily applied in downstream blood vessel characterization tasks such as the calculation of the length, tortuosity, or caliber of the blood vessels. Yet, it is still unclear if pre-trained CNNs can provide robust, unbiased, results on downstream tasks when applied to datasets that they were not trained on. Here, we focus on measuring the tortuosity of blood vessels and investigate to which extent CNNs may provide biased tortuosity values even after fine-tuning the network to the new dataset under study. We show that the tortuosity values obtained by a CNN trained from scratch on a dataset may not agree with those obtained by a fine-tuned network that was pre-trained on a dataset having different tortuosity statistics. In addition, we show that the improvement in segmentation performance when fine-tuning the network does not necessarily lead to a respective improvement on the estimation of the tortuosity. To mitigate the aforementioned issues, we propose the application of specific data augmentation techniques even in situations where they do not improve segmentation performance.

Eric K. Tokuda, Cesar H. Comin, Luciano da F. Costa

An important issue in clustering concerns the avoidance of false positives while searching for clusters. This work addressed this problem considering agglomerative methods, namely single, average, median, complete, centroid and Ward's approaches applied to unimodal and bimodal datasets obeying uniform, gaussian, exponential and power-law distributions. A model of clusters was also adopted, involving a higher density nucleus surrounded by a transition, followed by outliers. This paved the way to defining an objective means for identifying the clusters from dendrograms. The adopted model also allowed the relevance of the clusters to be quantified in terms of the height of their subtrees. The obtained results include the verification that many methods detect two clusters in unimodal data. The single-linkage method was found to be more resilient to false positives. Also, several methods detected clusters not corresponding directly to the nucleus. The possibility of identifying the type of distribution was also investigated.

Mayra Z. Rodriguez, Cesar H. Comin, Dalcimar Casanova et al.

Many real-world systems can be studied in terms of pattern recognition tasks, so that proper use (and understanding) of machine learning methods in practical applications becomes essential. While a myriad of classification methods have been proposed, there is no consensus on which methods are more suitable for a given dataset. As a consequence, it is important to comprehensively compare methods in many possible scenarios. In this context, we performed a systematic comparison of 7 well-known clustering methods available in the R language. In order to account for the many possible variations of data, we considered artificial datasets with several tunable properties (number of classes, separation between classes, etc). In addition, we also evaluated the sensitivity of the clustering methods with regard to their parameters configuration. The results revealed that, when considering the default configurations of the adopted methods, the spectral approach usually outperformed the other clustering algorithms. We also found that the default configuration of the adopted implementations was not accurate. In these cases, a simple approach based on random selection of parameters values proved to be a good alternative to improve the performance. All in all, the reported approach provides subsidies guiding the choice of clustering algorithms.

Cesar H. Comin, Thomas K. DM. Peron, Filipi N. Silva et al.

The increasing interest in complex networks research has been a consequence of several intrinsic features of this area, such as the generality of the approach to represent and model virtually any discrete system, and the incorporation of concepts and methods deriving from many areas, from statistical physics to sociology, which are often used in an independent way. Yet, for this same reason, it would be desirable to integrate these various aspects into a more coherent and organic framework, which would imply in several benefits normally allowed by the systematization in science, including the identification of new types of problems and the cross-fertilization between fields. More specifically, the identification of the main areas to which the concepts frequently used in complex networks can be applied paves the way to adopting and applying a larger set of concepts and methods deriving from those respective areas. Among the several areas that have been used in complex networks research, pattern recognition, optimization, linear algebra, and time series analysis seem to play a more basic and recurrent role. In the present manuscript, we propose a systematic way to integrate the concepts from these diverse areas regarding complex networks research. In order to do so, we start by grouping the multidisciplinary concepts into three main groups, namely features, similarity, and network connectivity. Then we show that several of the analysis and modeling approaches to complex networks can be thought as a composition of maps between these three groups, with emphasis on nine main types of mappings, which are presented and illustrated. Such a systematization of principles and approaches also provides an opportunity to review some of the most closely related works in the literature, which is also developed in this article.

D. R. Amancio, C. H. Comin, D. Casanova et al.

Pattern recognition techniques have been employed in a myriad of industrial, medical, commercial and academic applications. To tackle such a diversity of data, many techniques have been devised. However, despite the long tradition of pattern recognition research, there is no technique that yields the best classification in all scenarios. Therefore, the consideration of as many as possible techniques presents itself as an fundamental practice in applications aiming at high accuracy. Typical works comparing methods either emphasize the performance of a given algorithm in validation tests or systematically compare various algorithms, assuming that the practical use of these methods is done by experts. In many occasions, however, researchers have to deal with their practical classification tasks without an in-depth knowledge about the underlying mechanisms behind parameters. Actually, the adequate choice of classifiers and parameters alike in such practical circumstances constitutes a long-standing problem and is the subject of the current paper. We carried out a study on the performance of nine well-known classifiers implemented by the Weka framework and compared the dependence of the accuracy with their configuration parameter configurations. The analysis of performance with default parameters revealed that the k-nearest neighbors method exceeds by a large margin the other methods when high dimensional datasets are considered. When other configuration of parameters were allowed, we found that it is possible to improve the quality of SVM in more than 20% even if parameters are set randomly. Taken together, the investigation conducted in this paper suggests that, apart from the SVM implementation, Weka's default configuration of parameters provides an performance close the one achieved with the optimal configuration.