文章摘要

文章信息/Info

Title:: Research on intelligent facial pain assessment method based on deep learning

作者:: 刘定玺1; 2; 3; 4; 蒋国璋1; 2; 3; 章花4; 5; 胜照友1; 3; 刘康2; 3; 鲁辉4; 5; 刘融4; 5△; 1.武汉科技大学冶金装备及其控制教育部重点实验室，武汉 430081；2.武汉科技大学机械传动与制造工程湖北省重点实验室，武汉 430081；3.武汉科技大学精密制造研究院，武汉 430081；4.武汉科技大学附属普仁医院 医学创新与转化研究所，武汉 430081；5.武汉市普仁医院，武汉 430081

Author(s):: LIU Dingxil; 2; 3; 4; JIANG Guozhang1; 2; 3; ZHANG Hua4; 5; SHENG Zhaoyoul; 3; LIU Kang 2; 3; LU Hui4; 5; LIU Rong4; 5; 1.Key Laboratory of Metallurgical Equipment and Control Technology of Ministry of Education,Wuhan University of Science and Technology,Wuhan 430081，China；2.Hubei Key Laboratory of Mechanical Transmission and Manufacturing Engineering, Wuhan University of Science and Technology, Wuhan 430081；3.Precision Manufacturing Institute, Wuhan University of Science and Technology, Wuhan 430081；4.Institute of Medical Innovation and Transformation，Wuhan 430081；5.Department of Orthopadics, Wuhan Puren Hospital, Wuhan 430081

Keywords:: Pain intensity assessment; Convolutional neural network; Face image; Image preprocessing; Image augmentation; Expression database; Classifier design

摘要:: 为了提高疼痛评估的快捷性、直观性、连续性、准确性，本研究建立了一种基于深度学习和面部表情图像的疼痛强度智能评估模型，为准确评估疼痛病情、适时调整治疗方案、实施有效镇痛措施提供补充依据。该方法采集7位志愿者在6种不同疼痛强度状态下的表情图片，通过预处理和数据增广形成数据库，设计深度卷积神经网络并进行训练，使用特征级差异进行面部疼痛表情分类。在JAFFE、CK+数据库上进行模型性能验证和评价。模型在训练集、测试集分类准确率分别达到98.15%、90.91%，训练速度19 ms/step。模型在JAFFE、CK+数据库上的验证效果良好。结果表明，该方法在疼痛表情识别中准确有效，评分结果具有显著性。

Abstract:: In order to improve the rapidity, intuitions, continuity and accuracy of pain assessment, we established an intelligent pain intensity assessment model based on deep learning and facial expression images, so as to provide supplementary basis for accurate assessment of pain condition, timely adjustment of treatment plan and implementation of effective analgesic measures. The facial images of 7 volunteers in 6 different pain intensity states were collected and a database through preprocessing and data augmentation was formed. Deep convolution neural network was designed and trained, and feature level differences were used to classify facial pain expressions. The model performance was verified and evaluated on JAFFE and CK+ database. The classification accuracy of the model in training set and test set reached 98.15% and 90.91% respectively, the training speed was 19 ms/step. The model is validated well on JAFFE and CK+ database. The results show that the model is accurate and effective in the recognition of pain expressions and the scoring results are significant.