Estimation for pain from facial expression based on XQEDA and deep learning

Accurately measuring pain from facial expressions is crucial in medicine for diagnosis, treatment, and drug testing. Traditional methods rely on patient feedback, which can be subjective and unreliable, especially for those who cannot communicate, like stroke patients or infants. This highlights the need for automated methods to objectively assess pain. Our research introduces two novel fusion structures to improve pain measurement accuracy. First, we use convolutional neural networks (CNNs) to estimate different pain levels from facial images. Second, we developed a multi-tensor variant of the XQDA algorithm, called XQEDA, to assess pain intensities. Previous methods primarily used facial expressions with mixed results, but our approach, including the new Tensor XQEDA (eXtended Quadratic Exponential Discriminant Analysis), aims to provide a more accurate solution. Our technique achieved a hight classification accuracy on the UNBC-McMaster dataset, demonstrating its effectiveness over other state-of-the-art methods.