AI-Assisted Forensic Analysis of Hanging-Related Ligature Marks: A Pilot Study Using Convolutional Neural Networks

Area of Science:

• Forensic pathology
• Medical image analysis
• Artificial intelligence

Background:

• AI application in forensic pathology is limited.
• Assessing ligature marks in hanging deaths is challenging for forensic experts.
• This study pilots a deep learning approach for classifying these marks.

Purpose of the Study:

• To evaluate a deep learning approach for morphological classification of hanging-related ligature marks.
• To establish a framework for AI-assisted forensic image analysis.

Main Methods:

• A Convolutional Neural Network (CNN) was trained on 404 standardized JPEG images.
• The dataset included hanging-related ligature marks, strangulation, and post-mortem artefacts.
• The model was validated internally and tested on independent case images from Italy and Lithuania.

Main Results:

• The CNN achieved an F1-score of 0.81 ± 0.04 in distinguishing hanging-related ligature marks.
• The study established criteria for image standardization in AI-assisted forensic analysis.
• The model showed encouraging performance in classifying lesions.

Conclusions:

• AI-based image analysis can potentially assist in evaluating ligature marks during external examinations.
• Forensic diagnosis necessitates integrating AI findings with autopsy, physical examination, and circumstantial evidence.
• Further research with larger datasets and multicenter protocols is required to confirm reliability and applicability.