Estudo e classificação de gestos de sinalização cirúrgica por meio de eletromiografia de superfície

• Ano de defesa: 2022
• Autor(a) principal: Freitas, Melissa La Banca
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Ponta Grossa; Brasil; Programa de Pós-Graduação em Engenharia Elétrica; UTFPR
• Palavras-chave em Português: Eletromiografia; Instrumentos e aparelhos cirúrgicos; Classificação; Processamento de sinais; Electromyography; Surgical instruments and apparatus; Classification; Signal processing; CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA; Engenharia/Tecnologia/Gestão
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/30511

Hand Signals in surgery are a specific language based on the set of gestures adopted in a surgical environment to facilitate communication for procedures handling. Through the signaling, the surgeon transmits the information to the scrub nurse which instrument he needs, thus reducing the total time of the surgery and the errors caused by failures in verbal communication. With the advancement of technology, telesurgery systems and robotic surgery are increasingly present, and these surgical procedures need to be adequate in the face of technological advances. In this sense, the present work aims to present a feasibility analysis of the use of surface electromyography (sEMG) signals and pattern recognition tools to classify surgical signaling gestures, aiming to assist surgical procedures. The acquisition was made using the commercial MyoTM armband. The composition of the database involved the acquisition of data from 10 volunteers when performing 14 hand signals referring to the request for instruments (compress, thread on spool, untied thread, Backhaus forceps, hemostatic forceps, Kelly hemostat forceps, Farabeuf retractor, scalpel, needle holder, Doyen valve, Allis forceps, anatomical forceps, rat tooth forceps and scissors) in 30 acquisitions. Seventeen features were extracted from the time domain and two from the frequency domain. The selection of groups of characteristics was based on studies that also use the classification of gestures involving the hand and wrist and that presented good results in the classification. The classification of the 14 gestures was performed using the algorithms LDA (Linear Discriminant Analysis), QDA (Quadratic Discriminant Analysis), KNN (K-Nearest Neighbor), RF (Random Forests), SVM (Support Vector Machine), MLP (Multilayer Perceptron), and Ensemble (combination of classifiers). The proposed system is viable if there is a calibration before use for each individual and adjustment of the gestures used, as an accuracy of more than 90% was obtained for individuals with the best results, in addition to a 100% of accuracy hit rate for the 10 individuals in a given gesture.