Currently, neonatal pain assessment varies among health professionals, leading to late intervention and flimsy treatment of pain in several occasions. Therefore, it is essential to understand the deficiencies of the current pattern of pain assessment tools in order to develop new ones, less subjective and susceptible to external variable influences. The aim of this paper is to investigate neonatal pain assessment using two models of Deep Learning: Neonatal Convolutional Neural Network trained end-to-end and ResNet trained using Transfer Learning. We used for training two distinct databases (COPE and Unifesp) and our results showed that the use of multi-racial databases might improve the evaluation of automatic models of neonatal pain assessment.
Redes Adversárias Generativas (GANs) estão sendo cada vez mais usadas para gerar artificialmente vários tipos de dados. O treinamento dessas redes requer um conjunto de dados suficientemente grande e se torna um desafio com pequenos conjuntos. Trabalhos recentes propuseram novas abordagens para o treinamento de GANs com poucas amostras. Este trabalho analisa a distribuição espacial dos dados reais e sintéticos desses conjuntos, construindo subespaços de forma aleatória e variando o nível de espalhamento. Para variar o nível de espalhamento, este trabalho propõe o algoritmo k-Amostras Esparsas. Os resultados mostraram que pequenos conjuntos com uma distribuição espacial mais espalhada tendem a gerar dados com mais diversidade.
Neonatal pain assessment might suffer variation among health professionals, leading to late intervention and flimsy treatment of pain in several occasions. Therefore, it is essential to develop computational tools of pain assessment, less subjective and susceptible to external variable influences. Deep learning models, especially Convolutional Neural Networks, have gained ground in the last decade, due to many successful applications in image analysis, object recognitions and human emotion recognitions. In this context, the general aim this dissertation was analyse quantitatively and qualitatively models of Convolutional Neural Networks in the task neonatal pain classification through a computacional framework based in face images of two distinct databases (an international, named COPE, and other national, named UNIFESP). How specific aims were implemented, evaluated and compared the performance of three existent models used in literature: Neonatal Convolutional Neural Network (N-CNN) and two type of ResNet50 models. The quantitative results showed the excellence of N-CNN to neonatal pain assessment automatic, with average accuracy of 87.2% and 78.7% for the databases COPE and UNIFESP, respectively. However, the quantitative analysis showed that all neural models evaluated, including N-CNN models, can learn artifacts from the imagens and not variation discriminating in faces, thus showed the necessity more studies to apply this models in clinical practice
A avaliação da dor neonatal pode sofrer variações entre profissionais de saúde, resultando em intervenção tardia e tratamento inconsistente da dor. Neste contexto, o objetivo desta dissertação foi analisar, quantitativa e qualitativamente, modelos de Redes Neurais Convolucionais na tarefa de classificação automática da dor neonatal por meio de imagens de faces de dois bancos de dados distintos (um internacional, denominado COPE, e outro nacional, denominado UNIFESP). Os resultados quantitativos mostraram a superioridade da arquitetura N-CNN para avaliação automática da dor neonatal, com acurácias médias de 87.2% e 78.7% para os bancos de imagens COPE e UNIFESP, respectivamente. No entanto, a análise qualitativa evidenciou que todos os modelos neurais avaliados, incluindo a arquitetura Neonatal Convolutional Neural Network (N-CNN), podem aprender artefatos da imagem e não variações discriminantes das faces, mostrando a necessidade de mais estudos para aplicação de tais modelos na prática clínica em questão.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
customersupport@researchsolutions.com
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.