Jianglong Yan’s research while affiliated with University of São Paulo and other places

In this work, we present a network-based technique for chest X-ray image classification to help the diagnosis and prognosis of patients with COVID-19. From visual inspection, we perceive that healthy and COVID-19 chest radiographic images present different levels of geometric complexity. Therefore, we apply fractal dimension and quadtree as feature extractors to characterize such differences. Moreover, real-world datasets often present complex patterns, which are hardly handled by only the physical features of the data (such as similarity, distance, or distribution). This issue is addressed by complex networks, which are suitable tools for characterizing data patterns and capturing spatial, topological, and functional relationships in data. Specifically, we propose a new approach combining complexity measures and complex networks to provide a modified high-level classification technique to be applied to COVID-19 chest radiographic image classification. The computational results on the Kaggle COVID-19 Radiography Database show that the proposed method can obtain high classification accuracy on X-ray images, being competitive with state-of-the-art classification techniques. Lastly, a set of network measures is evaluated according to their potential in distinguishing the network classes, which resulted in the choice of communicability measure. We expect that the present work will make significant contributions to machine learning at the semantic level and to combat COVID-19.

Complex machine learning tasks for knowledge discovery in networked data, such as community detection, node categorization, network visualization, and dimension reduction, have been successfully addressed by coarsening algorithms. It iteratively reduces the original network into a hierarchy of smaller networks, resulting in informative simplifications of the original network at various degrees of detail. Few of these algorithms, however, have been specially built to cope with bipartite networks. Besides of this, current coarsening algorithms present the following theoretical limitations that should be addressed: 1) A high-cost search strategy in dense networks; 2) current coarsening algorithms are usually based on label propagation, which is limited to propagate single-labels; and 3) the synchronous label propagation scheme yields the cyclic oscillation problem. To overcome such limitations, we propose a coarsening algorithm based on multi-label propagation, which is more suitable for large-scale bipartite networks and allows a time-effective implementation. Furthermore, our proposal improves the standard semi-synchronous strategy and simultaneously propagates multiple labels to create the coarsened network representation. The empirical analysis of synthetic and real-world networks provides evidence that our coarsening strategy leads to significant gains regarding accuracy and runtime against standard techniques.

Traditional classification techniques usually classify data samples according to the physical organization, such as similarity, distance, and distribution, of the data features, which lack a general and explicit mechanism to represent data classes with semantic data patterns. Therefore, the incorporation of data pattern formation in classification is still a challenge problem. Meanwhile, data classification techniques can only work well when data features present high level of similarity in the feature space within each class. Such a hypothesis is not always satisfied, since, in real-world applications, we frequently encounter the following situation: On one hand, the data samples of some classes (usually representing the normal cases) present well defined patterns; on the other hand, the data features of other classes (usually representing abnormal classes) present large variance, i.e., low similarity within each class. Such a situation makes data classification a difficult task. In this paper, we present a novel solution to deal with the above mentioned problems based on the mesostructure of a complex network, built from the original data set. Specifically, we construct a core–periphery network from the training data set in such way that the normal class is represented by the core sub-network and the abnormal class is characterized by the peripheral sub-network. The testing data sample is classified to the core class if it gets a high coreness value; otherwise, it is classified to the periphery class. The proposed method is tested on an artificial data set and then applied to classify x-ray images for COVID-19 diagnosis, which presents high classification precision. In this way, we introduce a novel method to describe data pattern of the data “without pattern” through a network approach, contributing to the general solution of classification.

An important task in combating COVID-19 involves the quick and correct diagnosis of patients, which is not only critical to the patient’s prognosis, but can also help to optimize the configuration of hospital resources. This work aims to classify chest radiographic images to help the diagnosis and prognosis of patients with COVID-19. In comparison to images of healthy lungs, chest images infected by COVID-19 present geometrical deformations, like the formation of filaments. Therefore, fractal dimension is applied here to characterize the levels of complexity of COVID-19 images. Moreover, real data often contains complex patterns beyond physical features. Complex networks are suitable tools for characterizing data patterns due to their ability to capture the spatial, topological and functional relationship between the data. Therefore, a complex network-based high-level data classification technique, capable of capturing data patterns, is modified and applied to chest radiographic image classification. Experimental results show that the proposed method can obtain high classification precision on X-ray images. Still in this work, a comparative study between the proposed method and the state-of-the-art classification techniques is also carried out. The results show that the performance of the proposed method is competitive. We hope that the present work generates relevant contributions to combat COVID-19.

In real world data classification tasks, we always face the situations where the data samples of the normal cases present a well defined pattern and the features of abnormal data samples vary from one to another, i.e., do not show a regular pattern. Up to now, the general data classification hypothesis requires the data features within each class to present a certain level of similarity. Therefore, such real situations violate the classic classification condition and make it a hard task. In this paper, we present a novel solution for this kind of problems through a network approach. Specifically, we construct a core-periphery network from the training data set in such way that core node set is formed by the normal data samples and peripheral node set contains the abnormal samples of the training data set. The classification is made by checking the coreness of the testing data samples. The proposed method is applied to classify radiographic image for COVID-19 diagnosis. Computer simulations show promising results of the method. The main contribution is to introduce a general scheme to characterize pattern formation of the data “without pattern”.

Several coarsening algorithms have been developed as a powerful strategy to deal with difficult machine learning problems represented by large-scale networks, including, network visualization, trajectory mining, community detection and dimension reduction. It iteratively reduces the original network into a hierarchy of gradually smaller informative representations. However, few of these algorithms have been specifically designed to deal with bipartite networks and they still face theoretical limitations that need to be explored. Specifically, a recently introduced algorithm, called MLPb, is based on a synchronous label propagation strategy. In spite of an interesting approach, it presents the following two problems: 1) A high-cost search strategy in dense networks and 2) the cyclic oscillation problem yielded by the synchronous propagation scheme. In this paper, we address these issues and propose a novel fast coarsening algorithm more suitable for large-scale bipartite networks. Our proposal introduces a semi-synchronous strategy via cross-propagation, which allows a time-effective implementation and deeply reduces the oscillation phenomenon. The empirical analysis in both synthetic networks and real-world networks shows that our coarsening strategy outperforms previous approaches regarding accuracy and runtime.