3D-AI BehaviorAtlas

• Behavioral products are usually designed for specific behaviors, and the equipment is specialized accordingly, such as water mazes, three-chamber social behaviors, etc. These are highly standardized and require strict adherence to the paradigm to induce animal behavior step by step in specific scenarios.
• On one hand, BehaviorAtlas can integrate some traditional behavioral paradigms, as long as these devices allow the equipment box to be transparent and can be placed inside our collection equipment. This enables BehaviorAtlas to achieve more precise, three-dimensional animal movement trajectories and further analyze some induced behaviors.
• On the other hand, BehaviorAtlas is mainly used to analyze spontaneous behaviors, which are unspecified paradigms, where animals are simply placed in an open field to move freely. Evaluation metrics can include kinematic parameters of animal movement in the open field and patterns of animal sequences.
• Kinematic parameters refer to various body coordinate data obtained based on 3D tracking, calculating parameters such as displacement velocity, motion intensity, body length, height above ground, body angle, etc., which can amount to nearly 40 parameters. Based on these parameters, one can assess locomotor ability, establish models for locomotor injuries, and abnormal limb diseases such as Parkinson’s, cerebral hemorrhage, and spinal cord injury.
• Action sequences refer to categories of animal actions obtained by decomposing and clustering 3D trajectory sequences in short time intervals (with time resolutions reaching the second level), such as Grooming, Sniffing, Walking, Rearing, etc. This captures all the behaviors of animals throughout the experiment (also known as “behavioral sequencing” or behavior spectrum). Based on behavior spectrum, one can assess whether animals exhibit stereotyped or abnormal behaviors, analyze the temporal status of animal behavior, and establish biomarkers from a behavioral perspective.

• Calculation rule for movement distance: Utilize the animal’s center point to calculate the movement distance frame by frame, with the unit being millimeters (mm). The body’s center point is the average of the coordinates on the x and y axes of all body points. Then, based on this average coordinate, calculate the Euclidean distance of movement frame by frame.
• The motion distance is stored in the FrameLevel_paras/Paras_data directory of the H5 file as frame-by-frame motion distance.
• In routine analysis, the average value of the movement distance is calculated. If you require the total movement distance, you can simply sum up the frame-by-frame movement distances.
• Slight jitter of body points has minimal impact since the data has already been filtered and processed.

The BehaviorAtlas 3D-AI fine behavior analysis technology decomposes animal action subtypes using an unsupervised learning strategy, which means that the resulting action subtypes cannot be predefined with their meanings.

To understand the specific meanings corresponding to each class of behavior numbers, manual annotation is required. This involves watching each clipped video and providing textual descriptions based on experience. This annotation may be relatively straightforward for some classes, such as Walking, Running, Rearing, Grooming, Sniffing, etc. However, there are also more complex actions, some of which are uncommon, and others may involve multiple actions simultaneously. In such cases, descriptions must be made using words or phrases, such as “Running with jump preparation”.