Artificial intelligence for human gunshot wound classification.

Area of Science:

• Forensic Pathology
• Medical Imaging
• Artificial Intelligence

Background:

• Identifying gunshot entrance and exit wounds can be challenging due to subtle features.
• Deep learning shows potential for automated medical image classification tasks.

Purpose of the Study:

• To assess the feasibility of using a deep learning model to classify entrance and exit gunshot wounds in digital images.
• To evaluate the accuracy of AI in distinguishing between these wound types.

Main Methods:

• A ConvNext Tiny deep learning model was trained on 2418 images of gunshot wounds.
• The model was trained using the Fastai library with a 70/30 train/validation split.
• Performance was evaluated on a holdout set of 708 images.

Main Results:

• The deep learning model achieved 87.99% accuracy on the holdout set.
• Precision was 83.99%, recall 87.71%, and F1-score 85.81%.
• The model correctly classified 88.19% of entrance and 87.71% of exit wounds.

Conclusions:

• Deep learning models can accurately discern entrance and exit gunshot wounds from digital images.
• AI performance is comparable to that of forensic pathologists.
• This represents a novel application of AI in forensic pathology for wound analysis.