Automated measurement of stress scores in video recordings of laboratory animals

在2591的第一个资助期内，参与小组分析了一些客观评估动物实验严重性的方法，以确定它们是否适合作为应激标志。FOR2591发现，基于图像和行为的分数，如老鼠鬼脸等级(MGS)，显示出与实际痛苦高度相关。传统上，这些分数是人工计算的，这导致了很高的手动工作量和潜在的偏见。为了加快和简化这一评估，成像和计算机视觉研究所一直致力于DFG资助的项目“实验室小鼠视频记录中压力分数的自动测量”，该项目一直与For 2591有关。项目的目标是证明评估基于图像的严重程度评分的自动化方法的一般能力，这一目标已经成功实现。我们能够证明这一过程可以自动化，并可以显著减少兽医人员所需的评分时间。在即将到来的资助期，我们的目标是利用正在进行的项目中获得的见解来开发和验证一个系统，用于对家中的老鼠和大鼠进行长期监控。这些动物将使用2D和3D摄像头组合进行永久监控，并将获取行为参数。与For 2591合作伙伴合作，我们将开发一种记录装置，扩展实验室动物科学中常用的家笼类型，并可以毫不费力地集成到现有的实验室环境中。随后，我们将开发定位、识别和分析动物的算法，以便评估集体收容所中单个动物的状况。为此，我们将分析运动信息，如旅行距离和位置偏好，以及个人行为，如养育或自我打扮，以及社会行为，如跟随或避开其他群体成员。将对项目合作伙伴进行的动物实验导致的这些行为的变化进行分析，以揭示干预措施和由此导致的行为变化之间的统计相关性。我们将利用成熟的图像处理方法以及人工智能和神经网络的最新进展。这些方法将与For 2591中选定的合作伙伴一起进行评估，并随后与在多中心研究中与小鼠和大鼠合作的所有小组一起进行验证。这样，我们将评估所开发的方法的普适性。

During the first FOR 2591 funding period, the participating groups have analysed a number of methods for objective assessment of severity in animal experiments regarding their suitability as stress markers. The FOR 2591 has found out that image- and behaviour-based scores such as the mouse grimace scale (MGS) show a high correlation with the actual distress. Traditionally, these scores are taken manually, which results in high manual workload and potential bias. To speed up and simplify this assessment, the Institute of Imaging and Computer Vision has been working on the DFG-funded project “Automated measurement of stress scores in video recordings of laboratory mice” , which has been associated with the FOR 2591. The project goal of proving the general capability of automated methods for assessing image-based severity scores has been successfully reached. We were able to demonstrate that this process can be automated and can lead to a measurable reduction in scoring time required by veterinary staff.In the upcoming funding period, we aim at using the insights gained during the ongoing project to develop and validate a system for long-term monitoring of mice and rats in their home cages. The animals will be permanently monitored utilizing a combination of 2D and 3D cameras and behaviour parameters will be acquired. In cooperation with FOR 2591 partners, we will develop a recording setup that extends home cage types that are commonly used in laboratory animal science and can be effortlessly integrated into existing lab environments. Subsequently, we will develop algorithms for localization, identification and analysis of animals to allow assessment of the status of individual animals in group housings. To this end, we will analyse locomotor information such as travelled distance and location preference as well as individual behaviour such as rearing or self-grooming and social behaviour such as following or avoiding other group members. Changes to these behaviours as a consequence of animal experiments conducted by the project partners will be analysed to uncover statistical correlations between interventions and the resulting changes in behaviour. We will utilize established image processing methods as well as current advances from artificial intelligence and neural networks. The methods will be evaluated with selected partners from the FOR 2591 and subsequently validated together with all groups who work with mice and rats in multicentre studies. This way, we will assess the generalizability of the developed methods.