An experimental study of objective pain measurement using pupillary response based on genetic algorithm and artificial neural network
An experimental study of objective pain measurement using pupillary response based on genetic algorithm and artificial neural network
复制标题
DOI：
10.1007/s10489-021-02458-4
复制
发表时间：
2021-05-17
影响因子：
5.3
通讯作者：
Lin, Yingzi
中科院分区：
文献类型：
--
作者：
Wang, Li;Guo, Yikang;Lin, Yingzi
获得患者疼痛水平的客观测量对于医疗保健提供者来说一直是一个挑战。在医院环境中最常见的疼痛评估方法是询问患者的口头评分，这被认为是一种主观方法。为了获得患者的客观疼痛水平，我们提出使用瞳孔反应和机器学习算法客观地测量疼痛水平。本研究在东北大学招募了32名健康受试者。通过要求健康受试者将手放入装满冰水的桶中，对他们施加疼痛刺激。我们从瞳孔直径数据中提取了11个特征。为了得到最优的特征子集，使用遗传算法（GA）来选择人工神经网络（ANN）分类器的特征。在特征选择之前，所有11个特征的ANN的f1得分为54.0 +/- 0.25%。在特征选择之后，使用所选择的特征子集，即均值、均方根（RMS）和瞳孔曲线下面积（PAUC），ANN具有最佳性能，准确率为81.0%。实验结果表明，瞳孔反应与机器学习算法一起可能是一种有前途的客观疼痛水平评估方法。这项研究的结果可以改善患者在远程医疗中的疼痛测量体验，特别是在大流行期间，大多数人不得不呆在家里。
Obtaining an objective measurement of the pain level of a patient has always been challenging for health care providers. The most common method of pain assessment in the hospital setting is asking the patients' verbal ratings, which is considered to be a subjective approach. In order to get an objective pain level of a patient, we propose measuring pain level objectively using the pupillary response and machine learning algorithms. Thirty-two healthy subjects were enrolled in this study at Northeastern University. A painful stimulus was applied to healthy subjects by asking them to place their hands inside a bucket filled with iced water. We extracted 11 features from the pupil diameter data. To get the optimal subset of the features, a genetic algorithm (GA) was used to select features for the artificial neural network (ANN) classifier. Before feature selection, the f1-score of ANN was 54.0 +/- 0.25% with all 11 features. After feature selection, ANN had the best performance with an accuracy of 81.0% using the selected feature subset, namely the Mean, the Root Mean Square (RMS), and the Pupillary Area Under Curve (PAUC). The experimental results suggested that pupillary response together with machine learning algorithms could be a promising method of objective pain level assessment. The outcomes of this study could improve patients' experience of pain measurement in telehealthcare, especially during a pandemic when most people had to stay at home.