Developing an Artificial Intelligence Chatbot for Snake Image Classification and Accuracy Improvement.

Area of Science:

• * Artificial Intelligence
• * Herpetology
• * Medical Informatics

Background:

• * Snakebites present a substantial global health burden, necessitating rapid and precise snake species identification for effective antivenom selection.
• * Accurate species identification is critical for guiding appropriate clinical management following snakebite incidents.

Purpose of the Study:

• * To assess the efficacy of a deep learning model, specifically Swin Transformer v2, for classifying snake species in Taiwan using real-world image data.
• * To evaluate the impact of a novel preprocessing technique, test-time object detection and cropping (TT-ODC), on model performance with challenging external image datasets.

Main Methods:

• * Development of a deep learning model based on Swin Transformer v2 architecture, utilizing transfer learning on a large dataset of labeled snake images.
• * Training and validation using 12,000 images from a 30,573 image dataset. Evaluation on an external test set of 2,400 images collected via a LINE chatbot and social media.
• * Implementation and assessment of the TT-ODC preprocessing method on the external test set to enhance image quality and model accuracy.

Main Results:

• * The Swin Transformer v2 model achieved 95.6% accuracy on the internal validation set.
• * Performance on the external test set without preprocessing was 83.3%.
• * Application of TT-ODC preprocessing boosted external test set accuracy to 89.8%, nearing human expert performance (90.3%).

Conclusions:

• * Integrating the Swin Transformer v2 model with the TT-ODC preprocessing method offers a practical and accurate solution for snake species identification in clinical settings.
• * This approach significantly enhances the reliability of AI-driven snake identification tools, particularly when dealing with images from diverse, real-world sources.
• * The developed system provides a valuable tool for improving snakebite management and patient outcomes.