Abstract:

Objective In traditional laboratory animal centers, there are issues such as low cage scheduling efficiency, insufficient supervision of personnel behavior, and difficulty in upgrading aging equipment. This study aims to upgrade existing laboratory animal centers by applying multimodal large-model technology. This upgrade intends to achieve real-time perception of the status of animal cages, intelligent supervision of experimental personnel behavior, and automated management of business workflows, thereby improving management efficiency and refinement. Methods This study proposes an artificial-intelligence-based informatization approach for upgrading laboratory animal centers, which can be compatible with different breeding equipment. The system architecture, from the bottom up, consists of three layers: the hardware facilities layer, the core algorithm layer, and the application functions layer. The hardware facilities layer is equipped with cameras and high-speed network transmission devices for collecting information on cages and personnel. The core algorithm layer utilizes multi-stage image preprocessing technology and multimodal large-model recognition technology to extract and identify image information. The application functions layer integrates the recognition results with the existing information of the animal center to generate real-time cage occupancy heatmaps, which can visually and clearly show the density distribution of cage usage in the laboratory animal center. Results The artificial-intelligence-based management system has achieved a cage recognition accuracy of 98.5% and a correct identification rate of laboratory coats of 98.8%. The average image processing time is 3.7 seconds per image, the effective utilization rate of cages has increased by 23%, and the turnover efficiency has improved by 35%. In addition, the management system can track and warn against non-compliant behaviors in real time. After intelligent recognition, the system has detected more violations, with the violation detection rate rising by 90.6%. After continuous use for three months, the weekly average number of violations decreased by 54.0% compared to the baseline period. Conclusion This study innovatively applies multimodal large models to the field of laboratory animal management, achieving real-time monitoring and automated management of cage marking, thereby improving management efficiency and precision. The system integrates multi-source data such as visual recognition and behavior analysis, establishing a comprehensive intelligent supervision system for experimental personnel. It provides research institutions with efficient, accurate, and cost-effective management tools, promoting the intelligent development of laboratory animal management.

Key words: Experimental animals, Large language model, Intelligent recognition, Information system