Deep Learning-Based Diagnosis of Fatal Hypothermia Using Post-Mortem Computed Tomography.

Area of Science:

• Forensic Medicine
• Radiology
• Artificial Intelligence

Background:

• Fatal hypothermia diagnosis presents challenges in forensic medicine due to non-specific findings, particularly in trauma cases.
• Post-mortem computed tomography (PMCT) aids diagnosis, but subtle hypothermia indicators in images are difficult for inexperienced pathologists to discern.

Purpose of the Study:

• To develop and evaluate a deep learning-based system for diagnosing fatal hypothermia.
• To assess the feasibility of this AI system as an alternative diagnostic aid for forensic pathologists.

Main Methods:

• Development of a deep learning model using an in-house dataset of forensic autopsy-proven fatal hypothermia cases.
• Performance evaluation of the deep learning system using metrics including the area under the receiver operating characteristic curve (AUC), sensitivity, and specificity.

Main Results:

• The deep learning system achieved a human-expert comparable AUC value of 0.905.
• The system demonstrated high diagnostic performance with a sensitivity of 0.948 and a specificity of 0.741.

Conclusions:

• The developed deep learning system is a useful and feasible tool for diagnosing fatal hypothermia.
• This AI-driven approach can potentially assist forensic pathologists in identifying subtle signs of fatal hypothermia on PMCT scans.