Context-Aware Mouse Behavior Recognition Using Hidden Markov Models.

Area of Science:

• Neuroscience
• Computational Biology
• Machine Learning

Background:

• Automated recognition of mouse behaviors is vital for studying psychiatric and neurologic diseases.
• Analyzing the temporal dynamics of mouse behaviors, especially rapid transitions, is challenging but essential.
• Existing methods struggle with visually similar behaviors and imbalanced datasets.

Purpose of the Study:

• To develop and implement a novel hidden Markov model (HMM) algorithm for describing temporal characteristics of mouse behaviors.
• To propose a hybrid deep learning architecture for enhanced mouse behavior recognition.
• To improve the accuracy and robustness of automated mouse behavior analysis.

Main Methods:

• Developed a novel hidden Markov model (HMM) algorithm tailored for temporal mouse behavior analysis.
• Proposed a hybrid deep learning architecture combining unsupervised and supervised layers.
• Utilized spatial-temporal segment Fisher vector encoding for visual and contextual features in the unsupervised layer.
• Employed a segment aggregate network in supervised layers to estimate HMM state-dependent observation probabilities.

Main Results:

• The proposed architecture effectively discriminates between visually similar mouse behaviors.
• Achieved high recognition rates, demonstrating the model's efficacy.
• Successfully processed imbalanced mouse behavior datasets, a common challenge in the field.
• Outperformed other state-of-the-art approaches on JHuang's and custom datasets.

Conclusions:

• The novel hybrid deep learning HMM algorithm provides a robust and accurate method for automated mouse behavior recognition.
• This approach enhances the analysis of temporal dynamics in mouse behaviors, aiding research in neurologic and psychiatric diseases.
• The method's ability to handle imbalanced data and visually similar behaviors marks a significant advancement in the field.