Automated Personal Identification Using Single CT Slices and Computer Vision

Overview

This study demonstrates the feasibility of automated personal identification using single cranial CT slices processed with a computer vision (CV) algorithm. By extracting distinctive image features from specific midface regions, the method achieved reliable matching against an antemortem CV database, simplifying identification in emergency and forensic contexts.

Background

Unknown individuals frequently present challenges in emergency medicine and forensic investigations, especially when no direct identity clues exist. Computed tomography (CT) is routinely used for diagnosis and virtual autopsies, offering detailed anatomical information. Previous identification methods using paranasal sinuses and CT required expert segmentation or convolutional neural networks, limiting automation. This study extends CV-based identification from orthopantomograms (OPGs) to single CT slices, aiming to automate and simplify personal identification.

Data Highlights

Key Findings

• Single CT slices from six defined midface regions can be used for automated personal identification using CV feature extraction.
• The AKAZE algorithm effectively detects robust keypoints and descriptors in CT images, resilient to rotation, scale, and lighting variations.
• Image preprocessing steps including Sobel edge enhancement and noise reduction improve feature detection.
• Matching the most recent CT examination against an antemortem CV database enables reliable identification when the highest scoring match corresponds to the individual.
• Artifacts such as missing teeth or metal implants can hinder feature extraction and identification accuracy.
• The method does not require manual segmentation or CNN training, facilitating automation and potential clinical application.

Clinical Implications

Automated personal identification using single CT slices can expedite identification processes in emergency and forensic settings, especially when traditional methods are limited. The approach leverages existing CT imaging protocols without additional scanning or complex manual processing, potentially improving workflow efficiency. Awareness of artifacts that may impair identification is important for interpreting results.

Conclusion

This study validates a novel, automated CV-based method for personal identification from single CT slices, demonstrating its feasibility and potential utility in clinical and forensic practice. The approach simplifies identification workflows by eliminating the need for expert segmentation or extensive training data.

References

1. Jena University Hospital IRB 2019 -- Study Approval and Protocol
2. AKAZE Algorithm -- Feature Extraction Method
3. Previous CV Identification Studies [11,12,13]