Currently, the medical care provided in a wide range of clinical applications makes use of the latest technologies, multimedia data, multidimensional medical images, videos, and data based on sensors and texts [1]. In the area of mental health, algorithms and data mining techniques are used to extract large volumes of predictive data, build models based on large-scale medical information, and predict diagnoses [2]. Despite its advantages in the medical field, these techniques and algorithms pose a computational challenge. Complex data management and software design are necessary [3,4]. Medical applications linked to the network sciences, which are responsible for collecting, managing, analyzing, interpreting, and presenting data, allow the development of visualization tools that support diagnosis and medical decision making [5,6].

Today, we have web-based 3D visualization tools that allow the exploration and analysis of data [7,8]. BeGraph is a computing application, in the cloud, dedicated to the visualization and analysis of complex networks for companies and the scientific field. Networks are becoming popular in the field of data analysis and big data because they contain information that other data structures lack [9]. Characteristics, such as the phenomenon of the small world, the diffusion of events, and the grouping in clusters based on communities, are exclusive of networks. Additionally, networks are the basic data structure in BeGraph. A network is a set of N entities called nodes related by a set of L links that represent binary relationships. Nodes and links can have associated properties, either imported from a data source or calculated with BeGraph [10].

The BeGraph software uses different metrics that consist of mathematical operations based on the graphics theory used to characterize the network. These metrics provide information about the centrality of the nodes, busiest links, cohesion properties, grouping of networks, and topology [11,12].

The purpose of this research was to analyze the trend of mental health illnesses in a region of Spain through BeGraph in order to make health-related decisions, thus helping medical personnel to focus their research on preventing and improving the life quality of patients most affected by the most common diseases.

There are similar studies that show the viability of our research. In a previous study [1], the authors presented a 3D medical graphical avatar (MGA). It is a web-based personal health record visualization system designed to explore web-based delivery of a wide array of medical data types including multidimensional medical images, medical videos, text-based data, and spatial annotations. In a previous report [7], the authors developed an interactive 3D tool to facilitate the visualization and exploration of covariate distributions and imbalances across evidence for network meta-analysis (NMA). In another report [13], researchers developed 3D molecular dynamics visualization (MDV) that offers an immersive viewing experience of biomolecular systems.

Below, we provide a description of BeGraph, the methodology used, the results obtained in the study, and the discussion and conclusions of the investigation.

In this section, we present the results using a database that included a total of 13,531 anonymous medical records. We used BeGraph software, which allows the user to upload, store, and share several networks in the cloud. Each network can be viewed and explored using a web browser, giving a general idea of its structure and topological properties [10].

To obtain the results shown below, we used different metrics and one of the BeGraph design algorithms (Force Atlas Layout) [14]. It is an algorithm based on a system directed by force and optimized to deal with large networks, and it is highly configurable. Degree centrality is responsible for identifying the most popular nodes in the network. The size of the node and the warmth of the color represent the degree of each node. Eigenvector centrality is responsible for measuring the centrality of the selected node, normalizing the components of the eigenvector to a maximum value of 1.

With the 3D visualization tool, we explored the data and represented the results. We seek to clarify aspects such as (1) networks of diseases prevalent in the 2005-2015 period of patient admission records and (2) networks of prevalent diseases depending on the patient’s gender, with consideration of each of the health care complexes included in the study.

In Figure 1, the prevalent diseases of the University Clinic of Valladolid are shown during the period from 2005 to 2015.

The nodes are identified with a property called NodeLabel, the ubiquitous node property used to identify the nodes in metrics and filters. The largest nodes represent the year and the smallest the diseases detected in that hospital. The links represent the binary relationships between them. In Figures 1, 2, and 3, it can be seen that there are diseases that were only registered in a given year and not present in the rest of the years. The nodes of the diseases that are interrelated with more than 1 year represent the same type of disease registered in several years.

In Figures 2 and 3, the networks with prevalent diseases in the Health Care Complex of Zamora and the Rio Hortega Hospital, respectively, are shown. The period analyzed is from 2005 to 2015 and visually reflects the relationship of the data, providing medical personnel with criteria for decision making.

The results of the main diseases presented by patient admission records with mental health diseases in the different health care complexes are shown in Table 1.

The percentages are obtained from the total records. The University Clinic of Valladolid had 7118 records, Zamora Hospital had 4128 records, and Rio Hortega Hospital had 2285 records.

Another parameter analyzed in our study was prevalent diseases depending on the patient’s gender, as shown in Figures 4, 5, and 6.

In Figures 4 and 5, the main nodes represent the gender of the admission records. The upper node corresponds to men and the lower node corresponds to women. The small nodes represent the diseases detected in the University Clinic of Valladolid and Zamora Hospital, respectively. In the Rio Hortega Hospital, the upper node corresponds to women and the lower node corresponds to men (Figure 6). The diseases are visually shown only in men and women, and the common ones among them are displayed.

Tables 2 and 3 present a summary of the main diseases detected in each of the health care complexes, with consideration of gender.

The results presented in Tables 2 and Table 3 show considerable differences in the types of disorders. Men have disorders related to schizophrenia and alcohol, while women have dysthymic and adaptive disorders.

To obtain the different networks shown in Figures 1-6, we used the layout algorithm called Force Atlas, which considers the network as a system of N nodes on which different forces act. In the algorithm configuration, we used 500 iterations, gravity of 1, jitter tolerance of 0.5, and mass range of 10, while in the configuration of the metrics, the eigenvector centrality used 100 iterations. These parameters allowed us to obtain the best visualizations of the networks formed by the data.

In this study, we performed a visual analysis of mental health diseases prevalent in a region of Spain, using the BeGraph software, in order to make health-related decisions. The results show that BeGraph is a tool that allows visualization, in 3D, of the behavior of a network, offering complementary information to classical statistical analysis. As a novelty, we presented the analysis and visualization of medical data from the point of view of the theory of complex networks. Metrics based on networks or graphs provide new properties, which can be used to describe the data quantitatively or as additional properties in other machine learning algorithms. The visualization of the network provides an interactive method of scanning medical data as well as qualitative characterization quickly and easily.

In this work, we used a database of 13,531 medical records. The analysis allowed us to determine the main disease detected in each of the health care complexes. We noted 6.50% (463/7118) of admissions involving unspecified paranoid schizophrenia at the University Clinic of Valladolid, 9.62% (397/4128) of admissions involving chronic paranoid schizophrenia with acute exacerbation at the Zamora Hospital, and 8.84% (202/2285) of admissions involving dysthymic disorder at the Rio Hortega Hospital. In the health care complexes with the most admissions, such as the University Clinic of Valladolid and Zamora Hospital, the greatest mental health disorder was presented by patients with different types of schizophrenia. These data allowed us to focus on this disease to analyze its behavior and help in diagnosis and treatment. In Rio Hortega Hospital, dysthymic disorder represented the main disease. It was present in other health care complexes with similar amounts of admissions, but represented a lower value with regard to the main diseases detected in the University Clinic of Valladolid.

The analysis of the main diseases depended on gender. The results of the disorders were different in men and women. In the case of men, in the three health care complexes, the highest number of admissions detected was related to schizophrenia disorders and alcohol dependence. In women, the disorders were based on dysthymic disorders, adaptive disorders, schizoaffective disorders, and poisoning. Therefore, the results indicated that it is necessary to take gender into account in order to make decisions in diagnoses. The trends shown by these results provide a basis for future research regarding the prediction of mental health diseases and common patterns among patients with the same diseases. In this case, we focused on schizophrenia disorders taking into account gender, stay days, health care complex, age, and other diagnosed diseases. We propose the prediction of patient readmissions with schizophrenia using machine learning techniques.