Facial features detection system for classifying critical situation of brain surgery postoperative patients

Abstract

The purpose of this research is to create a model for detecting a help signal from stroke patients after surgery. Through artificial intelligence technology by detecting via a patient's face image to help reducing workload and hospital’s expenses. The data is collected from the Stroke Unit, King Chulalongkorn Memorial Hospital at the intensive care unit (ICU). The total number of samples is 8 persons, all these 8 were screened by the doctor and completed the consent forms before data being collected. The method of data collection is performed by attaching 3 small cameras on three positions: the left, right and top of the patient bed, to obtain a straight face image of the patient from every angle. The recording time is approximately 5-7 days after that the patient's image will bring into the data preparation step by selecting pictures of the patient who needs help from various facial features. There are two models to be chosen for comparison of efficiency: Logistic Regression Model, and Neural Network Model. The performance measurement, the Confusion Matrix was used to compare both models, including others well-known research. The result shows that the proposed neural network model has accuracy in prediction is 92%, the precision and recall are both equal to 92%.