Comparative Study of Deep Learning Algorithms for the Detection of Facial Paralysis

Statistical analysis has shown a noteworthy escalation in the rate of brain strokes in recent times. But the time required to carry out the conventional methods to detect brain stroke on a patient requires both trained personnel and time. Any discrepancy in the process might also lead to fatal consequences. Facial paralysis is among the most frequent signs of stroke. Currently, there are not many effective models for detecting facial paralysis. To help make the stroke detection technique less formidable to at least a certain extent, this report presents a comparative analysis of deep learning algorithms for the efficient detection of facial paralysis. We have compared three different pre-trained Convolutional neural networks (CNN) to find the one best suitable for detecting facial paralysis. The models include ResNet50, InceptionV3 and VGG-16. The models were trained using datasets from Kaggle consisting of facial images. Five different evaluation metrics, which are accuracy, precision, recall, F1-score, and area under the ROC (receiver operating characteristic) curve were used. We were able to attain the highest accuracy using ResNet50, which was 96.63%, while InceptionV3 yielded an accuracy of 95.38% and VGG-16 gave an accuracy of 93.50%. Overall, we can say that this research can aid in implementing models that can increase the efficiency and accuracy in the detection of facial paralysis in biomedical fields.