Deep learning and firearm wound classification: a pilot study

Area of Science:

• Forensic pathology
• Artificial intelligence
• Medical imaging analysis

Background:

• Deep learning (DL) applications in forensic medicine are emerging, particularly for visual analysis tasks.
• Previous studies demonstrated AI potential in predicting shooting distance from GSW images with high accuracy.
• Forensic pathology can benefit from AI tools to enhance visual analysis of evidence.

Purpose of the Study:

• To explore the application of deep learning techniques for classifying gunshot wounds (GSW).
• To evaluate the performance of AI in GSW pattern recognition compared to existing benchmarks.
• To assess the feasibility of AI in forensic pathology for wound analysis.

Main Methods:

• Utilized Lobe AI software for a 4-phase study: training, validation, testing, and data analysis.
• Classified GSWs into four categories: GSWs, entrance/exit wounds, range of fire, and ammunition type.
• Trained the AI model using images from a forensic atlas and tested with case history photos and intact skin controls.

Main Results:

• Observed encouraging data, with numerous parameters exceeding human performance thresholds.
• Achieved high predictive values in specific areas, surpassing previous scientific evidence.
• AI demonstrated strong potential in classifying GSW characteristics.

Conclusions:

• Limited data availability was an intrinsic limitation, necessitating further algorithm development.
• The study highlights the potential of AI in forensic pathology, with strengths in category analysis and control usage.
• Future multicenter research with larger sample sizes is crucial for developing a generalizable forensic AI model.