A deep learning-based system for automatic detection of emesis with high accuracy in Suncus murinus

Area of Science:

• Computational Biology
• Neuroscience
• Pharmacology

Background:

• Traditional methods for quantifying emesis in Suncus murinus are labor-intensive and prone to operator error.
• Advancements in deep learning offer opportunities for automated and accurate animal behavior analysis.

Purpose of the Study:

• To develop and validate an Automatic Emesis Detection (AED) tool for precise quantification of emesis in S. murinus.
• To leverage deep learning, specifically 3D convolutional neural networks and self-attention mechanisms, for automated behavior analysis.

Main Methods:

• Utilized three-dimensional convolutional neural networks and self-attention mechanisms to create the AED tool.
• Trained the model using motion-induced emesis videos from S. murinus.
• Validated the AED tool's performance across various emetics, including resiniferatoxin, nicotine, and cisplatin.

Main Results:

• Achieved an overall accuracy of 98.92% for emesis detection in S. murinus.
• Demonstrated high accuracy for motion-induced emesis (99.42%) and across various emetic agents (ranging from 96.93% to 100%).

Conclusions:

• Deep learning-based automatic analysis enhances efficiency, accuracy, and reduces human bias in quantifying animal behavior.
• The AED tool offers valuable insights for developing automated behavioral analysis models in S. murinus, supporting preclinical research and drug development.