Navegando por Palavras-chave "Artificial Intelligence"

Agora exibindo 1 - 11 de 11
  • Imagem de Miniatura
    Item
    Acesso aberto (Open Access)
    Aplicação de redes neurais artificiais para o diagnóstico de arritmias cardíacas usando processamento de sinais de eletrocardiograma.
    (Universidade Federal de São Paulo, 2022-10-14) Lima, Victor Mendes Cunha [UNIFESP]; Santos, Sergio Ronaldo Barros dos [UNIFESP]; http://lattes.cnpq.br/0608523738367987; http://lattes.cnpq.br/9949464804738379
    Este trabalho trata da aplicação de redes neurais artificiais (RNAs) como método de diagnóstico de arritmias cardíacas, a exemplo de infarto no miocárdio, utilizando técnicas de processamento de sinais de eletrocardiograma obtidos a partir de um conjunto de dados com informações de diagnóstico de pacientes saudáveis e de pacientes com diversas patologias cardíacas. Os dados utilizados para treinamento e teste das RNAs foram obtidos a partir da extração dos espectros das derivações dos sinais de eletrocardiograma. Em seguida os espectros obtidos foram filtrados e, a partir deles, foram calculadas as potências espectrais em sete bandas de frequências distintas pelo método de Burg. Após a obtenção das potências espectrais fez-se a separação dos dados em grupo de treinamento e grupo de teste, usados para, respectivamente, treinar e testar as redes. Os treinamentos foram realizados por meio do algoritmo backpropagation e do algoritmo LMS. A rede com melhor desempenho apresentou taxa de acerto de 80,97%, enquanto a de pior desempenho apresentou taxa de acerto de 78,94%. Mesmo com valores abaixo do esperado, de acordo com outras literaturas, o método apresentou resultados próximos com diferentes separações de dados, revelando boa precisão.
  • Nenhuma Miniatura disponível
    Item
    Somente Metadadados
    Aplicação de técnicas de inteligência artificial ao desenvolvimento de um sistema de apoio a decisão para doença celíaca
    (Universidade Federal de São Paulo (UNIFESP), 2011) Tenório, Josceli Maria [UNIFESP]; Marin, Heimar de Fátima [UNIFESP]
    Introdução: o diagnóstico da doença celíaca é um processo complexo devido à multiplicidade dos sintomas, sinais, grupos de risco, formas de apresentação e intersecção dos sintomas com outras doenças. Para a confirmação da suspeita diagnóstica, é imprescindível a realização da biopsia do intestino delgado, o padrão-ouro. Objetivo: desenvolver um sistema de apoio à decisão, em ambiente web, integrado a um classificador automático para reconhecimento dos casos de doença celíaca. Métodos: um sistema web foi construído para suportar um protocolo eletrônico esquematizado para atendimento e registro dos dados clínicos dos pacientes. Uma avaliação preliminar de usabilidade foi realizada. Uma base de dados de retrospectiva com 178 casos clínicos para treinamento foi construída. Foram testados 270 classificadores automáticos disponíveis no software Weka 3.6.1, utilizando cinco técnicas de inteligência artificial, a saber, árvores de decisão, classificador bayesiano, k-vizinhos próximos, máquinas de vetor de suporte e redes neurais artificiais. As métricas analisadas foram área sob a curva ROC, sensibilidade, especificidade e taxa de acerto, utilizadas nessa sequência como critério para seleção do algoritmo a ser implantado no sistema web. O algoritmo com maior AUC foi selecionado e acoplado ao sistema web, gerando o software intitulado SADCEL. Uma base de dados de teste foi construída, com 38 casos clínicos, para a avaliação do SADCEL em relação à utilidade diagnóstica. A hipótese diagnóstica apontada pelo SADCEL foi comparada às indicadas pelos especialistas durante a realização da consulta por meio de estatística kappa. Resultados: o sistema web foi avaliado pelos usuários com nível excelente de usabilidade, com SUS-score de 83,5 ± 10,0. Na fase de treinamento, as melhores métricas foram apresentadas pelo algoritmo AODE F-1, do tipo classificador bayesiano, com taxa de acerto 80,0%, sensibilidade 0,78, especificidade 0,80 e AUC 0,84. Comparado ao padrão ouro, o SADCEL alcançou uma precisão de 84,2% com um nível de concordância diagnóstica de k = 0,68 (p <0,0001), o que indicou um bom nível de concordância. A mesma taxa de acerto foi obtida na comparação entre as indicações do diagnóstico dos especialistas e o padrão-ouro, com k = 0,64 (p-value <0,0001). Entre a indicação do especialista e do SADC, obteve-se k = 0,46 (p-value = 0,0008), o que indica concordância moderada. Conclusão: o nível de precisão alcançado pelo algoritmo de classificação automática integrado ao sistema web evidencia a utilidade potencial da SADCEL no auxílio ao diagnóstico de doença celíaca..
  • Imagem de Miniatura
    Item
    Acesso aberto (Open Access)
    Aplicação de técnicas de inteligência artificial ao desenvolvimento de um sistema de apoio à decisão para doença celíaca
    (Universidade Federal de São Paulo (UNIFESP), 2011-02-22) Tenório, Josceli Maria [UNIFESP]; Marin, Heimar de Fatima [UNIFESP]; Universidade Federal de São Paulo (UNIFESP)
    Introduction: the diagnosing of celiac disease involves some complexity due to its multiple symptoms, signs, risk groups, presentation and the wide possibility of differential diagnosis. In order to confirm the diagnosis of celiac disease, it is required to perform the biopsy or the small intestine, the gold standard. Objective: to develop a decision making support system, in web environment, including an automated classifier to recognize cases of celiac disease, to be previously selected among experimental models drawing upon techniques of artificial intelligence. Methods: a web system was implemented to support an electronic protocol designed to help with celiac disease investigation and collect clinical data. A preliminary assessment of this system usability was performed through the analysis of a questionnaire based on the System Usability Scale (SUS) completed by 10 direct users of the web system implemented. A retrospective database with 178 cases was build for training the automated classifier. A total of 270 automated classifiers available in the software Weka 3.6.1 were tested using 5 artificial intelligence techniques – decision tree, K-nearest-neighbor, Bayesian classifier, support vector machine and artificial neural networks. The parameters area under the receiver operating characteristic curve (AUC), sensitivity, specificity and correctness rate were used, in the order above, as criteria to select the classification algorithm to be implemented in the web system. The algorithm with the largest AUC was included in the web system whose software was named SADCEL. A database with 38 clinical cases was built to assess the diagnostic power this software. The diagnostic hypothesis obtained from SADCEL was compared with those reached by the specialists participating in the study using Kappa Statistic. Results: the preliminary usability score attained by the web system was 83.5 ± 10.0 (excellent). The Bayesian classifying algorithm AODE F1 had the best performance scoring 80.0% for correctness, 0.78 for sensitivity, 0.84 for specificity and 0.84 for AUC. Compared with the study gold standard, SADCEL achieved an accuracy of 84.2% with a level of agreement with the diagnostic gold standard rated as k = 0.68 (p-value < 0.0001), indicative of good level of agreement. The level of agreement between the specialist diagnostic hypothesis and the diagnostic gold standard was rated as k = 0.64 (p-value < 0.0001). The agreement between the specialist and SADCEL diagnostic hypotheses was rated as k = 0.46 (p-value) indicative of moderate level of agreement. Conclusion: the level of accuracy attained by the classifying algorithm incorporated in this study´s web system evidences the potential usefulness of SADCEL in helping with diagnosing celiac disease in clinical set. This study is, thus, expected to be a contribution towards the establishing of a computational means of diagnosing the celiac disease.
  • Nenhuma Miniatura disponível
    Item
    Somente Metadadados
    Construção de algoritmos de Machine Learning na Radiologia
    (Universidade Federal de São Paulo (UNIFESP), 2020-09-17) Kitamura, Felipe Campos [UNIFESP]; Abdala, Nitamar [UNIFESP]; Universidade Federal de São Paulo
    Recent research in artificial intelligence has shown great potential to change radiology as we know it today. The tools to aid the radiological diagnosis can bring numerous benefits to the patients, radiologists and referring physicians. Despite the high expectations for this technology, the path to the creation of clinically useful and safe tools is a huge challenge that involves several aspects. In this work, we will address ethical, regulatory, technical and cultural considerations that need to be addressed to expand the scope of artificial intelligence algorithms in practice. Next, we present 7 projects developed by our group that address some of the challenges in the area: (1) the lack of reproducibility when reading exams, (2) the creation of optimized algorithms for each clinical problem, (3) the limitation to access large volumes of quality annotated data, (4) the lack of reproducibility of artificial intelligence researches, (5) the difficulty of integrating algorithms in medical practice, (6) errors in the registration of exams types and (7) the risk of exposure of sensitive patient information.
  • Nenhuma Miniatura disponível
    Item
    Somente Metadadados
    Desambiguação de sentidos de palavras por meio de aprendizado semissupervisionado e word embeddings
    (Universidade Federal de São Paulo (UNIFESP), 2020-01-27) Sousa, Samuel Bruno Da Silva [UNIFESP]; Berton, Lilian [UNIFESP]; Universidade Federal de São Paulo
    Words naturally present more than one meaning and ambiguity is a recurrent feature in natural languages. Consequently, the task of Word Sense Disambiguation (WSD) aims at defining which word sense is the most adequate in a given context by using computers. WSD is one of the main problems in the field of Natural Language Processing (NLP) since many other tasks, such as Machine Translation and Information Retrieval, may have their results enhanced by accurate disambiguation systems. To solve this problem, several Machine Learning (ML) approaches have been used, such as unsupervised, supervised, and semi-supervised learning. However, the lack of labeled data to train supervised algorithms made models which combine labeled and unlabeled data in the learning process appear as a potential solution. Additionally, a comparative study of semi-supervised learning (SSL) approaches for WSD was not done before, as well as the combined employment of SSL algorithms with efficient word representations known as word embeddings, which became popular in the literature of NLP. Hence, the main goal of this work concerns the investigation of the performance of several semi-supervised algorithms applied to the problem of WSD, using word embeddings as features. To do so, four graph-based SSL algorithms were compared to each other on the main benchmark datasets for WSD. In order to check the word embeddings influence on the final results of the algorithms, six different setups for the Word2Vec model were trained and employed. The experimental results show that SSL models present competitive performances against supervised approaches, reaching over 80% of F1 score when only 25% of labeled data are input. Furthermore, these algorithms have the advantage of avoiding a new training step to classify new words.
  • Nenhuma Miniatura disponível
    Item
    Somente Metadadados
    Elaboração de algoritmos de prevenção de indivíduos com possível presença de depressão decorrente de deficiência visual
    (Universidade Federal de São Paulo (UNIFESP), 2020-04-30) Choi, Stefano Neto Jai Hyun [UNIFESP]; Santos, Vagner Rogerio Dos [UNIFESP]; Universidade Federal de São Paulo
    Objective: Develop a solution of Artificial Intelligence named Chatbot capable to apply a test of presence of depression on patients under ophthalmologic treatment for a best clinical monitoring. Methods: By using BLiP®, a prototype of Chatbot was developed introducing in it a social-demographic questionnaire used by the Psychobiology Department of UNIFESP and a test of diagnosis of presence of depression translated and validated in Brazil. After developing the prototype, it was tested by different forms, as functional, structural and validation tests, to ensure its functionality without involving human beings. Results: The prototype of Chatbot presented a great flow of conversation applying the test of presence of depression with reliably, showing that it is possible to apply on patients with depression symptoms when compared to the manual test application. Conclusion: The prototype developed will be a potential applier of the test of presence of depression to previously diagnose the patient, collecting and sending information to the medical team to attendance and interfere when necessary.
  • Nenhuma Miniatura disponível
    Item
    Somente Metadadados
    Estudo termodinâmico e modelagem matemática do equilíbrio líquido-vapor de misturas envolvendo Diesel e Biodiesel
    (Universidade Federal de São Paulo (UNIFESP), 2020-03-02) Melo, Edivaldo Bernardino De [UNIFESP]; Falleiro, Rafael Mauricio Matricarde [UNIFESP]; Universidade Federal de São Paulo
    The use of biodiesel has been growing in recent years. In Brazil there are laws that influence the addition of biodiesel to diesel oil. Knowing Liquid-Vapor Balance (ELV) data from these mixtures is essential to obtain improvements in separation, transportation and storage processes. Currently, ELV data from mixtures formed by components of the diesel/biodiesel mixture are scarce. Thus, the objective of this work was to experimentally determine ELV data involving hydrocarbons and ethyl esters at low pressures, and to model these data, via thermodynamics and Artificial Neural Networks. The data were obtained through Differential Exploratory Calorimetry (CSD), a technique that provides accurate results; in a short time and with small amounts of samples. Binary systems: hexadecane + ethyl myristato; dodecylbenzene + ethyl oleate; dodecylbenzene + ethyl stearate; ethyl laurato + dodecylbenzene; were obtained via CSD at a pressure of 2.67 kPa. Since these are unpublished systems, the UNIFAC predictive model was used to predict the equilibrium curve, the highest mean relative deviation found was 1.83%. The UNIQUAC and NRTL models were adjusted to the ELV data, and their binary parameters were calculated. Os valores dos coeficientes de atividade para o sistema: hexadecano + miristato de etila foram próximos de 1,0, indicando um comportamento próximo do ideal. The systems: dodecylbenzene + ethyl oleate, dodecylbenzene + ethyl stearate and ethyl laurate + dodecylbenzene, the values obtained were of the order of 3.0 for some regions, indicating a greater deviation from the ideality of the liquid phase. Artificial Neural Networks were used to calculate the ELV temperature from the molar fraction of the most volatile component in the mixture, critical temperature, critical pressure, acentric factor and mass connectivity index of each substance. For this, experimental data of binary systems formed by the compounds were used: fatty acids, ethyl esters and hydrocarbons. Two Neural Models were obtained and used in the prediction of the ELV of the boiling temperature of the mixture. Neural Model I presented for the myristic systems of ethyl + palmitic acid; ethyl palmitate + sauric acid a mean relative deviation respectively in relation to the UNIFAC model. Neural Model II presented for the hexadecane + ethyl myristato systems; dodecylbenzene + palmitic acid ethyl stearate, an average relative deviation of 5.02% and 2.29% respectively in relation to the data obtained via CSD.
  • Nenhuma Miniatura disponível
    Item
    Somente Metadadados
    Fuzzy Lymphedema Assessment based on Clinical and Functional Criteria
    (Ieee, 2011-01-01) Vicentini, Patricia [UNIFESP]; Araujo, Ernesto [UNIFESP]; Perez, Maria Del Carmen Janeiro [UNIFESP]; Universidade Federal de São Paulo (UNIFESP)
    A fuzzy lymphedema clinical and functional assessment for classifying the risk of developing and its severity is proposed in this paper. Different from the previous approach where reversibility, skin infection, and skin changes were the input variables, this new fuzzy lymphedema clinical assessment takes into account the elements that compose the Brazilian Society of Lymphology (SBL). Such a SBL-based metric not only takes into account pre-clinical cases with risk of worsening and infective-degenerative local complications but also points out functional data of limb edema with the involvement of articular joints. The proposed approach includes both clinical (Pitting, Skin Changes, Stemmer Signal and Reversibility) and functional criteria (Joint Involvement). The fuzzy lymphedema assessment based on clinical and functional criteria allows establishing therapeutic global rehabilitative programs, degree of assistance necessity of patients, and reduction of the daily living activities.
  • Imagem de Miniatura
    Item
    Acesso aberto (Open Access)
    Modelo de aprendizado de máquina para predição de Diabetes tipo 2 por meio de variáveis de fácil acesso
    (Universidade Federal de São Paulo, 2023-03-10) Silva, Leonardo Fernandes [UNIFESP]; Caranti, Danielle Arisa [UNIFESP]; http://lattes.cnpq.br/4760019839583649; http://lattes.cnpq.br/7376085574661825; Universidade Federal de São Paulo (UNIFESP)
    Objetivo: Validar um modelo preditivo de diabetes do tipo 2 utilizando aprendizagem de máquina através de variáveis de fácil acesso e comparar os resultados dos bancos de dados VIGITEL e NHANES para validação da metodologia. Métodos: Após a seleção dos bancos de dados VIGITEL (2015) e NHANES (2014,15,16,17), foi aplicado critérios de inclusão e exclusão, aqueles que foram diagnosticados acima dos 30 anos e dados não faltantes, em cima dos indivíduos finais foi utilizado o método de balanceamento SMOTE para melhor aplicação dos algoritmos. Uma vez balanceado, foram aplicados os algoritmos “árvore de decisão”, “Floresta Aleatória” e “floresta de isolamento”. Resultados: O modelo de predição de diabetes tipo 2 apresentou melhor desempenho em todas as métricas em comparação com as outras duas doenças crônicas (dislipidemia e hipertensão arterial) no conjunto de dados do NHANES. No VIGITEL, o diabetes teve melhor desempenho em sensibilidade (73,25%) em comparação com as outras duas doenças, a hipertensão também teve alto desempenho em especificidade e acurácia (79,51% e 73,63%). Entre os dois conjuntos de dados, o NHANES teve melhor desempenho em todas as métricas em diabetes e hipertensão. Conclusões: O presente estudo apresentou evidências para a criação de um modelo preditivo através da utilização de aprendizagem de máquina para auxiliar no diagnóstico precoce de doenças crônicas através de variáveis de fácil acesso.
  • Imagem de Miniatura
    Item
    Acesso aberto (Open Access)
    Modelo e metodologia para o ensino de oftalmoscopia direta e sua aplicação no desenvolvimento de algoritmos para interpretação de imagens oftalmológicas
    (Universidade Federal de São Paulo (UNIFESP), 2021) Martins, Thiago Goncalves Dos Santos [UNIFESP]; Schor, Paulo [UNIFESP]; Universidade Federal de São Paulo
    Objective: Develop methods to improve eye care, with a new model and teaching methodology for the study of direct ophthalmoscopy and development of new technologies for data and image analysis. Method: After studying the irregular distribution of ophthalmologists in countries like Brazil and Portugal, a questionnaire was carried out with non-ophthalmologists to assess the level of confidence in the direct ophthalmoscopy exam. Next, the human eye model was made from physical calculations using cardboard paper with a black background, acrylic sphere, and plaster. The model was applied in the teaching of direct ophthalmoscopy and red reflex test. New image and data analysis technologies have been developed. An algorithm was developed for the evaluation of macula edema in fundus color photography and an image analysis and control program for toxoplasmosis chorioretinitis. Results: The results of the questionnaire showed that doctors feel less confident in the diagnosis through direct ophthalmoscopy. The model proved to be effective in teaching direct ophthalmoscopy and red reflex test. Its versatility allowed it to be used for teaching veterinary medicine students. The developed algorithm proved to be useful in the detection of edema in fundus color photography of diabetic patients and the image analysis program proved to be useful for the monitoring of patients with toxoplasmosis uveitis. Conclusion: Searching for alternatives to improve the population's ophthalmic service, a simple and low-cost model of the human eye was developed to be used in the teaching of direct ophthalmoscopy and red reflex test, which enabled the teaching and training of this technique, including adapted for the teaching of direct ophthalmoscopy in veterinary medicine students. This can be a teaching method easily adopted by any educational institution due to its low cost and effectiveness. The development of new data and image analysis technologies has proven to be useful alternatives for the diagnosis and monitoring of ophthalmic diseases in situations where we do not have adequate access to eye careObjective: Develop methods to improve eye care, with a new model and teaching methodology for the study of direct ophthalmoscopy and development of new technologies for data and image analysis. Method: After studying the irregular distribution of ophthalmologists in countries like Brazil and Portugal, a questionnaire was carried out with non-ophthalmologists to assess the level of confidence in the direct ophthalmoscopy exam. Next, the human eye model was made from physical calculations using cardboard paper with a black background, acrylic sphere, and plaster. The model was applied in the teaching of direct ophthalmoscopy and red reflex test. New image and data analysis technologies have been developed. An algorithm was developed for the evaluation of macula edema in fundus color photography and an image analysis and control program for toxoplasmosis chorioretinitis. Results: The results of the questionnaire showed that doctors feel less confident in the diagnosis through direct ophthalmoscopy. The model proved to be effective in teaching direct ophthalmoscopy and red reflex test. Its versatility allowed it to be used for teaching veterinary medicine students. The developed algorithm proved to be useful in the detection of edema in fundus color photography of diabetic patients and the image analysis program proved to be useful for the monitoring of patients with toxoplasmosis uveitis. Conclusion: Searching for alternatives to improve the population's ophthalmic service, a simple and low-cost model of the human eye was developed to be used in the teaching of direct ophthalmoscopy and red reflex test, which enabled the teaching and training of this technique, including adapted for the teaching of direct ophthalmoscopy in veterinary medicine students. This can be a teaching method easily adopted by any educational institution due to its low cost and effectiveness. The development of new data and image analysis technologies has proven to be useful alternatives for the diagnosis and monitoring of ophthalmic diseases in situations where we do not have adequate access to eye care.
  • Imagem de Miniatura
    Item
    Acesso aberto (Open Access)
    Uso de rotinas de aprendizado de máquina em prontuário eletrônico para apoio a diagnósticos de pacientes oftalmológicos
    (Universidade Federal de São Paulo (UNIFESP), 2021) Alves, Lucas De Oliveira Batista [UNIFESP]; Santos, Vagner Rogerio Dos [UNIFESP]; Universidade Federal de São Paulo
    Objective: To implement artificial intelligence routines through machine learning to construct diagnostic prediction models with data from electronic medical records of patients from the Department of Ophthalmology of Hospital São Paulo. Method: Preparation of a literature review of the main techniques and solutions of machine learning to use in electronic medical records, 1. extraction, treatment and analysis of data from medical records of the Department; 2. construction and analysis of vectorization models of related words in the context of the Database of Hospital São Paulo; 3. construction and validation of diagnostic prediction models. Results: The word vectorization models were able to capture the semantics of medical terms and enabled the construction of diagnostic prediction models, making the prediction model a great tool to assist health professionals. Conclusion: The machine learning models showed potential results to assist as diagnostic support tools of ophthalmologic patients.