Deep learning for behaviour classification in a preclinical brain injury model

Area of Science:

• Neuroscience
• Computational Biology
• Biomedical Engineering

Background:

• Early detection of traumatic brain injuries (TBI) significantly impacts patient prognosis and survival.
• Current automated TBI detection methods rely on limited clinical diagnostics and basic statistical analysis, lacking generalizability for large populations.
• Advanced deep learning architectures offer potential for extracting complex information from large datasets for improved TBI assessment.

Purpose of the Study:

• To explore a novel multiple input, convolutional neural network and long short-term memory (LSTM) integrated deep learning architecture for TBI detection.
• To investigate the effectiveness of this deep learning model in predicting brain injury using behavioral data from a preclinical murine model.
• To compare the proposed model's performance against traditional machine learning algorithms like support vector machines and random forest classifiers.

Main Methods:

• Utilized a home cage automated (HCA) system to record behavioral data (distance traveled, body temperature, social separation, movement) of mice post-TBI or sham intervention over 72 hours weekly for 5 weeks.
• Trained a deep learning model, incorporating convolutional neural network and LSTM, on HCA behavioral data to predict the presence of brain injury.
• Addressed class imbalance in the training data and employed leave-one-out cross-validation for model evaluation.

Main Results:

• The proposed deep learning model demonstrated superior performance in detecting the presence of brain trauma in mice compared to support vector machines, random forest classifiers, and feedforward neural networks.
• The model effectively utilized behavioral data captured by the HCA system for accurate TBI prediction.
• Strategies for handling class imbalance in the uninjured control group were explored and evaluated.

Conclusions:

• The developed multiple input, convolutional neural network and LSTM deep learning architecture shows significant promise for automated and accurate TBI detection in preclinical models.
• Home cage automated behavioral monitoring combined with advanced deep learning offers a scalable approach for studying TBI outcomes.
• This approach could potentially translate to improved diagnostic tools for traumatic brain injuries in clinical settings.