Automated identification of unknown decedents: matching postmortem CT images with clinical databases

Area of Science:

• Forensic radiology
• Medical imaging analysis
• Computer vision applications

Background:

• The maxillary sinus is a key anatomical landmark for forensic identification.
• Virtual autopsies using postmortem computed tomography (CT) offer a non-invasive method for examining deceased individuals.
• Computer vision (CV) techniques show potential for automated analysis of medical images.

Purpose of the Study:

• To assess the reliability of identifying unknown deceased individuals by matching postmortem CT images from virtual autopsies with antemortem CT scans from a clinical database using CV.
• To evaluate the accuracy of CV techniques in identifying individuals based on maxillary sinus morphology.

Main Methods:

• CT images of the maxillary sinus were extracted from ten virtual autopsies.
• These images were compared against a database of 853 antemortem CT examinations from 738 individuals.
• Image processing and CV feature extraction were performed on 60,255 antemortem CT slices to create a CV database.

Main Results:

• The study achieved identification rates of 50% at rank 1, 80% at rank 2, and 100% at rank 7 among 738 potential identities.
• Challenges included partial maxillary sinus views or the presence of injuries in antemortem scans.
• Maximizing CV matching points requires postmortem imaging to closely replicate antemortem standards.

Conclusions:

• Individual identification using postmortem CT from virtual autopsies combined with clinical databases is feasible.
• Adherence to clinical imaging standards during postmortem imaging is crucial for improving CV matching accuracy.
• This approach holds promise for forensic identification, particularly when antemortem data is available.