Proposta de sistema integrado para detecção de temperatura corporal e expressões faciais

• Ano de defesa: 2023
• Autor(a) principal: Jorge, Amanda Rosa Ferreira
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; Brasil; Programa de Pós-graduação em Engenharia Biomédica
• Palavras-chave em Português: Pacientes UTI; ICU patients; Temperatura corporal; Body temperature; Expressões faciais; Facial expressions; Sistema integrado; Integrated system; CNPQ::ENGENHARIAS; CNPQ::ENGENHARIAS::ENGENHARIA BIOMEDICA::BIOENGENHARIA::PROCESSAMENTO DE SINAIS BIOLOGICOS; Engenharia biomédica; Controle de temperatura; Expressão facial; Pacientes hospitalizados; Unidade de tratamento intensivo
• Link de acesso: https://repositorio.ufu.br/handle/123456789/38781; http://doi.org/10.14393/ufu.di.2023.333

Intensive care units or intensive care centers are specialized hospital areas that provide specific treatment and continuous monitoring of the physical condition of hospitalized patients. The vital signs mostly used for monitoring are the electrocardiogram, heart rate, blood pressure, blood oxygen saturation and body temperature. In addition to being caused by the physical exhaustion of the intensivist professional, the loss of apparently irrelevant information from the patient generates delay in diagnosis and inaccuracy in the execution of medical interventions. Studies indicate that the isolated analysis of vital signs does not expose the patient's real physiological state, it is argued that it is necessary to implement a "system of systems" in which the global analysis groups apparently secondary signs that can help to reveal important hidden physical conditions . For example, fever can affect up to 75% of ICU patients, causing agitation and delirium. It is also verified that more than 50% of hospitalized patients present reports of moderate to severe pain during intubation in a conscious state and at rest, which may generate an increase in cardiac workload. The point at which agitation, anxiety, pain, delirium and sedation have in common is the patient's potential facial expressions. Thus, in addition to abnormal body temperature being characterized as fever, the facial expressions of patients in intensive care units may be indications of the need for medical intervention. The current literature makes little progress on the monitoring of facial expression and body temperature without contact of patients in general, but with special emphasis on non-communicative patients. Although efficient, the state-of-the-art models on facial expressions are not robust, are designed for applications in offline systems and have a high algorithm training cost. Therefore, the main objective of the present study is to develop models for capturing body temperature and for classifying facial expressions in an automated and efficient way from data acquisition by RGB camera. The emotion classification model is based on the analysis of distances between specific points on the face. The model was developed to classify seven facial emotions: anger, disgust, happy, neutral, sad, surprise and fear. For feature selection, the genetic algorithm technique was used and for classification, support vector machines were used. The results demonstrate that the developed technique is able to perform the classification from data with average accuracies above 80% for all classes, using images from the WSEFEP, CK+ and JAFFE databases. Surprise, happy and neutral emotions showed the best accuracy. For non-contact body temperature prediction, images captured from a conventional camera were used, in the visible spectrum, in RGB. Regression models were used to estimate body temperature, based on three points on the face: cheeks, subclavicle and forehead. The results show that the estimated average temperature varies according to the region (cheeks, subclavicular and forehead) with R² from 0.79 to 0.94. Mean squared errors are in the range of 0.15-0.20 ºC. It is thus concluded that the developed techniques are able to predict with a good degree of accuracy the emotions and body temperature of the subjects, without contact, and can be integrated into a single RGB camera monitoring system focusing on bedridden patients and non-communicative.