Facial Palsy Detection Through Changes in Facial Muscle Functionality Using CNN Algorithm

Facial palsy is a common condition in the medical field that results from facial nerve disorders, leading to weakened facial motor functions. Detecting facial palsy typically involves observing the functional ability of facial muscles, a process currently performed manually by healthcare professionals relying on visual cues. However, manual detection is prone to errors, this needs for early and automated facial palsy detection. Overall, our research aims to improve the detection of facial palsy by leveraging deep learning techniques. By automating the detection process, we hope to provide a more accurate and efficient approach for early identification of this condition. In this research, we aim to develop a facial palsy detection system using the Convolutional Neural Network (CNN) algorithm. We utilize three datasets, namely CK+, JAFFE, and YouTube Facial Palsy, which are organized into normal and dysfunctional classes. The dataset is divided into training and testing data in an 80:20 ratio, and preprocessing techniques such as image resizing and grayscale conversion are applied. To find the optimal parameters for our model, we employ k-fold cross validation during the training process. The evaluation results show that the best model achieved optimal performance with the following parameters: 3 hidden layers, a fully connected layer (fc)-1 with a hidden size of 128, fc-2 with a hidden size of 16, and fc-3 with a hidden size of 32. The optimizer used was Adam, with a learning rate of 0.001, a batch size of 64, and trained for 10 epochs. This best model attained an average training accuracy of 0.958, testing accuracy of 1, recall of 1, and precision of 0.99.