Marker-less tracking system for multiple mice using Mask R-CNN.

Area of Science:

• Neuroscience
• Animal Behavior
• Computational Biology

Background:

• Accurate evaluation of mouse behavior is critical for pharmacological research.
• Current methods are limited to single-mouse tracking and typically require light conditions.
• Limitations stem from human observation constraints and experimental tool capabilities.

Purpose of the Study:

• To develop a novel, marker-less tracking method for multiple mice using deep learning and top-view videos.
• To overcome limitations of existing single-mouse and light-dependent behavioral analysis techniques.
• To enable comprehensive analysis of mouse social behavior, particularly for nocturnal species.

Main Methods:

• A stepwise deep learning approach was employed, including mouse contour detection, unique identifier (ID) assignment, and semi-automatic error correction.
• The Mask Regional Convolutional Neural Network (Mask R-CNN) was trained on manually annotated mouse contour data from open-field arena recordings.
• Mouse IDs were assigned by calculating inter-frame similarities, with subsequent correction steps to refine predictions.

Main Results:

• The developed method accurately tracked two to four mice in videos recorded under light conditions.
• The technique demonstrated efficacy in analyzing videos captured under dark conditions, suitable for nocturnal mice.
• Successful application to C57BL/6 mice behavior in an open-field arena was achieved.

Conclusions:

• This novel marker-less tracking technology accurately monitors multiple mice, enhancing behavioral analysis capabilities.
• The method extends behavioral observation to dark conditions, facilitating the study of nocturnal mouse sociality.
• This advancement offers a new paradigm for understanding mouse social interactions and supports pharmacological research.