基于多模态的贝叶斯网络疼痛识别方法

有效的疼痛管理对病人的治疗和护理至关重要，针对传统的单模态疼痛识别准确度低的问题，提出了一种基于多模态的贝叶斯网络(multimodal Bayesian network, MMBN)疼痛识别方法。首先利用互信息对多模态特征进行相关性判断，剔除冗余的特征向量，使得模型简洁；其次将多模态特征与贝叶斯网络结构的可扩展性相结合设计了一种基于多模态的BN结构，并建立疼痛识别模型；最后利用BN概率推理算法完成疼痛识别，并在UNBC-McMaster数据库上进行验证。实验结果表明：与传统基于单模态的疼痛识别方法相比较，MMBN方法利用多模态之间的信息互补性能够有效地提高疼痛识别准确度，为目前的疼痛识别与研究提供了一种新手段。

Pain Recognition Method Based on Multimodal Bayesian Network

Effective pain management is very important to the treatment and care of patients. Aiming at the problem of low accuracy of traditional single modal pain recognition, a pain recognition method based on multimodal Bayesian network (MMBN) was proposed. Firstly, used mutual information to judge the correlation of multimodal features and eliminated redundant feature vectors make the model concise; Secondly, a multimodal BN structure is designed by combining the multimodal features with the scalability of the Bayesian network structure, and a pain recognition model is constructed; Finally, the BN probabilistic reasoning algorithm was used to complete the pain recognition and it was verified on the UNBC-McMaster database. The experimental results show that by comparing with the traditional pain recognition methods based on single modal, the MMBN method can effectively improve the accuracy of pain recognition by using the information complementarity between multiple modalities, besides, providing a new method for current pain recognition and research.