Deep learning in forensic gunshot wound interpretation-a proof-of-concept study
Petteri Oura1, Alina Junno2,3, Juho-Antti Junno2,3,4
1Center for Life Course Health Research, Faculty of Medicine, University of Oulu, Oulu, Finland. petteri.oura@oulu.fi.
International Journal of Legal Medicine
|April 6, 2021
Summary
Deep learning models can accurately predict gunshot wound distance from images, offering a potential new tool for forensic pathology. This study shows high accuracy in classifying wound types, aiding forensic analysis.
Area of Science:
- Forensic Science
- Computer Science
- Pathology
Background:
- Deep learning applications are widespread in medicine but limited in forensic science.
- Forensic pathology, a visual field, could benefit from AI tools for wound interpretation.
- Predicting shooting distance from gunshot wound images is a challenging but valuable forensic task.
Purpose of the Study:
- To evaluate the potential of trained neural network architectures in predicting gunshot wound shooting distance from photographs.
- To test the hypothesis that deep learning can classify shooting distances based on visual wound characteristics.
- To establish a proof-of-concept for AI-driven gunshot wound analysis in forensic pathology.
Main Methods:
- A dataset of 204 gunshot wound images from piglet carcasses was created, categorized by shooting distance (negative controls, contact, close-range, distant).
- Neural network architectures were trained, validated, and tested using the AIDeveloper open-source software.
- A multilayer perceptron model (MLP_24_16_24) was identified as the most effective architecture.
Main Results:
- The trained MLP_24_16_24 model achieved a 98% testing accuracy in classifying shooting distances.
- The model correctly classified all negative controls, contact shots, and close-range shots.
- One distant shot was misclassified, indicating high overall performance.
Conclusions:
- Deep learning models demonstrate significant potential for assisting forensic pathologists in gunshot wound interpretation.
- This study provides initial evidence supporting the use of AI for determining shooting distance from wound images.
- Further large-scale research is encouraged to advance deep learning applications in forensic wound analysis.
