Abstract Summary/Description
Tracking maternal care behaviors in mice is essential for neuroscience research, particularly in understanding early-life stress. Traditional methods, such as manual tagging with BORIS, are resource-intensive and time-consuming, prompting the need for an automated solution. The objective of this project is to develop TailOR, an automated system that utilizes advanced computer vision techniques to tag specific maternal behaviors in mice models with minimal manual input. This system aims to streamline the process, improve tagging accuracy, and reduce human involvement. We employ SAM2 (Segment Anything Model 2) to generate masks for both the mouse and the nest in each frame, allowing for precise tracking of behaviors. By calculating the centroids of these masks, we monitor displacements across frames to identify patterns of motion. DeepLabCut is used for pose detection, and behavior classification is conducted through a Neural Network based on spatial data. Preliminary prototyping efforts and evaluation reveal that it is possible to detect and track mouse activities with practically usable accuracy. The project is ongoing, and the computer vision algorithm for behavioral classification is under development. Key obstacles include extended training times, low contrast, and partial visibility of pups. TailOR represents a significant advancement in the automation of maternal care behavior analysis in mice, offering a scalable solution that reduces human labor and enhances the accuracy of behavioral studies. This technology has potential applications beyond neuroscience, benefiting any field that requires detailed, reliable behavior tracking in animal models.