Diagnóstico de doenças mentais baseado em mineração de dados e redes complexas

Abstract

O uso de técnicas de mineração de dados tem produzido resultados importantes em diversas áreas, tais como bioinformática, atividades de transações bancárias, auditorias de computadores relacionados à segurança, tráfego de redes, análise de textos, imagens e avaliação da qualidade em processos de fabricação. Em medicina, métodos de mineração de dados têm se revelado muito eficazes na realização de diagnósticos automáticos, ajudando na tomada de decisões por equipes médicas. Além do uso de mineração de dados, dados médicos podem ser representados por redes complexas, de modo a incluir conexões entre seus elementos. Por exemplo, no caso do cérebro, regiões corticais podem representar vértices em um grafo e as conexões podem ser definidas através das atividades corticais. Com isso, pode-se comparar a estrutura do cérebro de sujeitos sadios com a de pacientes que apresentam doenças mentais de modo a definir métodos para diagnóstico e obter conhecimento sobre como a estrutura do cérebro está relacionada com alterações comportamentais e neurológicas. Nesse trabalho, estamos interessados em usar métodos de mineração de dados e redes complexas para classificar pacientes portadores de quatro diferentes tipos de doenças mentais, isto é, esquizofrenia, autismo, deficit de atenção/desordem de hiperatividade e paralisia progressiva nuclear.Palavras-chave: Inteligência Artificial, Mineração de dados, Redes complexas, Doenças neuro- lógicas.A data mining and knowledge discovery is in a field of research, with applications in different areas such as bioinformatics, customer transaction activity, security related computer audits, network traffic, text analysis and quality evaluation in manufacturing. In medicine, data mining methods have proven very effective in performing automatic diagnostics, helping in making decisions by medical teams. In addition to the use of data mining, medical data can be represented by complex networks in order to include connections between its elements. For example, in the case of the brain, cortical regions can represent vertices in a graph and the connections can be defined through cortical activities. Thus, we can compare the brain structure of healthy patients with those of patients with mental disorder in order to define methods for diagnosis and to obtain knowledge about how the structure of the brain is related to behavioral and neurological changes. Here, we are interested in using data mining methods and complex networks to classify patients with four different types of mental desorders, that is, schizophrenia, autism, attention deficit / hyperactivity disorder, and progressive supranuclear paralysis.

Diagnosis of autism spectrum disorder based on functional brain networks and machine learning

“…It also checked whether the use of metrics was better than a direct use of time series -the one of better performance would be chosen. After the best brain connectivity metric had been determined, the following ML classifiers were used: Random Forest (RF) [59], Naive Bayes (NB) [60], Logistic regression (LG) [61] with L-BFGS (Limited-memory Broyden Fletcher Goldfarb Shanno) solver [62], Multilayer Perceptron (MLP) [63], and tuned convolution neural network (called here tuned CNN) implemented in [64]. SHAP value method was used for biological interpretation, since it allows to explain individual predictions of each attribute.…”

Analysis of functional connectivity using machine learning and deep learning in different data modalities from individuals with schizophrenia

“…After the best brain connectivity metric had been determined, the following ML classifiers were used: random forest (RF) [58], Naive Bayes [59], multilayer perceptron [60], tuned convolution neural network (called here CNN tuned and CNN untuned ) implemented in [61], and long short-term memory (LSTM) neural networks [62]. In addition to the CNN deep learning used in prior work [39], the LSTM network is a recurrent neural network commonly used to identify patterns in time series.…”

Analysis of functional connectivity using machine learning and deep learning in multimodal data from patients with schizophrenia

“…After the best brain connectivity metric had been determined, the following ML classifiers were used: Random Forest (RF) [56], Naive Bayes (NB) [57], Multilayer Perceptron (MLP) [58], tuned Convolution Neural Network (called here CN N tuned and CN N untuned ) implemented in [59], and Long Short-Term Memory neural networks (LSTM) [60]. In addition to the CNN deep learning used in prior work [37], the LSTM network is a form of recurrent neural network commonly used to identify patterns in time series.…”