Automated facial paralysis detection using deep learning / Nurul Natasha Razlan, Nurbaity Sabri and Raihah Aminuddin

Facial paralysis, stemming from nerve issues, results in the inability to control facial muscles, leading to asymmetry and weakness. This condition not only affects appearance but also disrupts daily activities. Diagnosis is time-consuming and requires specialised expertise and equipment. To address these challenges, a deep learning-based system is proposed to analyse facial expressions and distinguish between normal and paralyzed states. “Automated Facial Paralysis Detection using Deep Learning” system leveraging the InceptionResNetV2 model, undergoes pre processing, feature extraction, and feature classification. Facial images are pre processed with techniques like data augmentation for robustness. Features are extracted to identify relevant characteristics, which are then classified using InceptionResNetV2. Evaluation on a Kaggle dataset, divided into training, validation, and testing sets with a ratio of 5:1:1, shows an impressive accuracy of 92.7% in identifying normal and paralyzed facial expressions. This underscores InceptionResNetV2's unmatched effectiveness in facial paralysis detection, marking significant progress in healthcare diagnostics.