Pain analysis from mouse facial expression using machine learning

Abstract

Pain is a fundamental sensation to perceive tissue injury. Since various diseases induce tissue injury, many patients are suffering from it. Therefore, intensive research has been carried out using experimental animal models to investigate the mechanism and to develop the effective therapy. Recent methods for pain assessment like grimace scale scoring and von Frey test depend on researcher’s manual observation. These human-powered tests are laborious, time-consuming, and low-throughput, and lack in the objectivity and reproducibility. The technology of artificial intelligence especially neural network achieved a remarkable progress. Among them, convolutional neural network (CNN) has been the de facto standard method for image recognition tasks. We here established an automated pain assessment method from the face image of mice using CNN. CNN trained with hundreds of thousands face images could accurately predict “pain” or “no pain” from a face image (sensitivity: 97%, specificity: 99%). We also revealed that trained CNN was applicable for the assessment of pain killer. In this section, I would like to introduce the detailed method, results, and application of our methods.