Automate facial paralysis detection using vgg architectures

Abstract

Facial Paralysis (FP) is a debilitating condition that affects individuals worldwide by impairing their ability to control facial muscles and resulting in significant physical and emotional challenges. Precise and prompt identification of FP is crucial for appropriate medical intervention and treatment. With the advancements in deep learning techniques, specifically Convolutional Neural Networks (CNNs), there has been growing interest in utilising these models for automated FP detection. This paper investigates the effectiveness of CNN architectures to identify patients with facial paralysis. The proposed method leveraged the depth and simplicity of Visual Geometry Group (VGG) architectures to capture the intricate relationships within facial images and accurately classify individuals with FP on the YouTube Facial Palsy (YFP) dataset. The dataset consists of 2000 images categorised into individuals with FP and non-injured individuals. Data augmentation techniques were used to improve the robustness and generalisation of the approach proposed. The proposed model consists of a features extraction module utilising the VGG network and a classification module with a Softmax classifier. The performance evaluation metrics include accuracy, recall, precision and F1-score. Experimental results demonstrate that the VGG16 model scored an accuracy of 88.47% with a recall of 83.55%, precision of 92.15% and F1-score of 87.64%. The VGG19 model attained level of precision of 81.95%, with a recall of 72.44%, precision of 88.58% and F1-score of 79.70%. The VGG16 model outperformed the VGG19 model in terms of accuracy, recall, precision, and F1-score. The results indicate that VGG architectures are effective in identifying patients with facial paralysis.