Intelligent system for infants' pain detection: pain intensity estimation using deep learning approach

Pain detection is an important agent for good pain management, especially for patients who are unable to express pain verbally as infants. Recently, professionals have depended on traditional assessment tools to detect pain but these tools have many limitations that may lead to poor pain management. For that, many researchers intended to find approaches to detect pain without these limitations, and the Artificial Intelligence (AI) field is the best for that. In this study, we proposed a deep learning model to estimate different levels of the pain intensity of full-term infants in the range (0-9) based on facial expressions that had been recorded during daily medical procedures in the NICU. We built a regression CNN model with a transfer learning technique and used a pre-trained VGG16 model with fine-tuning to improve the performance of classification and avoid overfitting. The model yielded a good performance with 0.494 MAE and 0.435 MSE. This study can contribute to accurately detecting the pain for infants which assists in achieving effective pain management. And make the system portable and user-friendly by embedding the model into a web application.