Bonnichsen Rikke, Nielsen Glenn Gunner Brink, Dam Jeppe Seidelin, Schrøder-Petersen Dorte, Buchmann Kurt
Danish Technological Institute, Taastrup, Denmark.
Department of Veterinary and Animal Sciences, University of Copenhagen, Frederiksberg C, Denmark.
J Fish Dis. 2025 Jan;48(1):e14027. doi: 10.1111/jfd.14027. Epub 2024 Sep 30.
A novel video-based real-time system based on AI (artificial intelligence) was developed to detect clinical signs in fish exposed to pathogens. We selected a White Spot Disease model involving rainbow trout as the experimental animal and the parasitic ciliate Ichthyophthirius multifiliis as a pathogen. We compared two identical fish tank systems: one tank was infected by co-habitation, whereas the other tank was kept non-infected (sham infection). The two fish tanks were separately video monitored (full top and side view) during the course of infection, during which fish were removed whenever they developed clinical signs (direct visual inspection by the observer). Image analysis (object detection, classification and tracking) was used to track behavioural changes in fish (in every recorded video frame), focusing on movement patterns and spatial localisation. Initially, the two fish groups (infected and non-infected) exhibited similar behaviour and non-infected fish did not change behaviour during the 15 d observation period (from 5 d before infection until 10 dpi). At 4, 7, 8, 9 and 10 dpi some infected fish showed clinical signs (equilibrium disturbance, gasping and lethargy) and were removed from the experiment. Anorexia occurred from 5 dpi and a gradual progression of gasping behaviour was noted, whereas the frequency of fish flashing (rubbing/scratching against objects) was low. Equilibrium disturbances and the development of white spots in the skin appeared to be a much later (8-10 dpi at this temperature) indicator of infection. The video analysis showed a general distribution of non-infected fish in all parts of the fish tank during the entire experiment, whereas infected fish already at 4-5 dpi moved towards higher water currents in the top and bottom positions. This change of fish positioning within the tank appeared as a promising early indicator of infection. The study suggests that continuous monitoring of fish behaviour using AI can potentially optimise the timing of humane endpoints, indicate disease signs earlier and thereby improve animal welfare in both animal experimentation and in aquaculture settings.
开发了一种基于人工智能（AI）的新型视频实时系统，用于检测接触病原体的鱼类的临床症状。我们选择了一个以虹鳟鱼为实验动物、寄生纤毛虫多子小瓜虫为病原体的白点病模型。我们比较了两个相同的鱼缸系统：一个鱼缸通过同居感染，而另一个鱼缸保持未感染（假感染）。在感染过程中，对两个鱼缸分别进行视频监测（顶部和侧面全景），在此期间，一旦鱼出现临床症状（由观察者直接目视检查），就将其取出。图像分析（目标检测、分类和跟踪）用于跟踪鱼的行为变化（在每个记录的视频帧中），重点关注运动模式和空间定位。最初，两组鱼（感染组和未感染组）表现出相似的行为，未感染的鱼在15天观察期内（从感染前5天到感染后10天）行为没有变化。在感染后4、7、8、9和10天，一些感染的鱼出现临床症状（平衡失调、喘息和嗜睡）并被从实验中取出。厌食从感染后5天开始出现，喘息行为逐渐加重，而鱼闪烁（摩擦/刮擦物体）的频率较低。平衡失调和皮肤出现白点似乎是感染的一个更晚出现的指标（在此温度下为感染后8 - 10天）。视频分析显示，在整个实验过程中，未感染的鱼在鱼缸的各个部位分布均匀，而感染的鱼在感染后4 - 5天就已经向鱼缸顶部和底部的较高水流处移动。鱼缸内鱼位置的这种变化似乎是感染的一个有希望的早期指标。该研究表明，使用人工智能持续监测鱼的行为可以潜在地优化人道终点的时间，更早地指示疾病迹象，从而在动物实验和水产养殖环境中改善动物福利。
