Dr. Duc Phong Nguyen  Researcher of the International Research Institute for Artificial Intelligence and Data Science (IAD), Dong A University. Duc Phong NGUYEN received his Ph.D. in Biomechanic Engineering at the University of Technology of Compiègne (France). His current research interest is machine learning, deep learning, reinforcement learning for facial recognition, and rehabilitation applied to medical computer-aided system diagnosis.

Title: A framework for facial rehabilitation based on Reinforcement Learning coupled with the finite element model of the face 

Abstract: 

Artificial intelligence (AI) is the study of computers and algorithms with the purpose of creating intelligent machines that can learn from data and make predictions. Biomechanics, on the other hand, is the study of mechanical principles relating to the movement and structure of living creatures. Computation is now less expensive, quicker, and more powerful. At the same time, a large amount of data, including photographs, text, and medical information, was available. The combination of AI and biomechanics has yielded significant results in the research of human movement mechanisms. This aids in the detection of numerous diseases in the human body, as well as the prevention of injuries, treatment decisions, and rehabilitation. On the other hand, facial palsy patients or patients under facial transplantation have facial dysfunctionalities and abnormal facial motion. This is due to the altered facial nerve and facial muscle systems. Knowing which muscle is responsible for what movement could support clinicians identify straightforward muscles that should be targeted for surgery. It, however, is almost practically hard or impossible to directly measure muscle activations from living subjects due to safety and accessibility limitations. Invert dynamics were frequently utilized to estimate muscle for and contraction using a biomechanical model. Any inverse dynamic solution, however, has a key restriction in that it is dependent on the availability of movement patterns as input. This information is not always easy to get from live people, particularly in the case of patients who have facial palsy or have had a face transplant. In the study, the facial motion learning capacity will be explored by the coupling between reinforcement learning and finite element modeling. The main objective is to provide, for the first time, the modeling workflow for this complex coupling and then to evaluate different learning strategies to establish motion patterns of the face during facial expression motions.