基于目标检测框架的大鼠面部疼痛快速自动评估方法  

A Fast and Automatic Method for Facial Pain Assessment in Rats Based on an Object Detection Framework

作　　者：陈璐炜 邓丁凡 李邦华 顾予佳 Chen Luwei;Deng Dingfan;Li Banghua;Gu Yujia(Biological Sciences and Medical Engineering Department of Southeast University,Nanjing 210096,China)

机构地区：[1]东南大学生物科学与医学工程学院,南京210096

出　　处：《信息化研究》2025年第2期23-28,共6页INFORMATIZATION RESEARCH

基　　金：国家自然科学基金项目(No.61902064)。

摘　　要：大鼠是疼痛相关研究中最常用的实验动物之一,如何准确评估大鼠的疼痛程度具有重要的意义。基于鬼脸量表的疼痛评估方法可以和人工智能技术深度结合,以往的实验动物智能化面部疼痛评估具有二阶段任务较为繁琐、仍需人工参与的缺陷,无法达到实用级别。本文利用深度学习技术,提出一种基于目标检测框架的大鼠面部疼痛快速自动评估方法:首先使用任意的目标检测算法进行训练和预测,得到初步鼠脸检测和疼痛评估结果,再根据“平均-归并”准则筛选得到每张图像最终有效的疼痛评估结果。该方法利用了目标检测算法的特点将鼠脸检测和疼痛评估转化为一阶段任务,无需人工参与;并且对目标检测算法会出现的漏检、误检现象进行设计处理,保证了预测结果的准确性。为验证所提出方法的性能,以3种目标检测算法为例,在RatsPain数据库上进行了基于留一大鼠协议的实验。实验结果表明,本文所提出的方法不仅能够进行快速的大鼠面部疼痛评估,而且也能达到较高的准确度,为大鼠的面部疼痛评估智能化提供了一种新思路。Rats are one of the most commonly used experimental animals in pain-related research,and accurately assessing their level of pain is of great significance.The pain assessment method based on the grimace scale can be deeply integrated with artificial intelligence technology.Previous intelligent facial pain assessments for experimental animals had the limitations of involving complex two-stage tasks and still requiring human intervention,preventing them from reaching practical levels.This study proposes a rapid and automatic method for facial pain assessment in rats based on an object detection framework by using deep learning technology.First,any object detection algorithm is used for training and prediction to obtain initial rat face detection and pain assessment results.Then,the final effective pain assessment result for each image is selected based on the“mean-merging”criterion.This method leverages the characteristics of object detection algorithms to transform rat face detection and pain assessment into a one-stage task without human involvement.Additionally,it addresses issues of missed and false detections that may occur in object detection algorithms,ensuring the accuracy of the predicted results.To verify the performance of the proposed method,experiments were conducted on the RatsPain database using a leave-one-rat-out protocol with three different object detection algorithms as examples.The experimental results demonstrate that the proposed method not only enables rapid facial pain assessment in rats but also achieves high accuracy,providing a new approach to intelligent rat facial pain assessment.

关 键 词：疼痛评估 大鼠 深度学习 鬼脸量表 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]