Clinical Scorecard: Single CT Image-Based Automated Personal Identification

At a Glance

Key Highlights

• Automated personal identification using single CT slices of defined craniofacial regions with computer vision algorithms
• Use of AKAZE algorithm for robust feature extraction resilient to rotation, scale, and lighting variations
• Retrospective study of 819 cranial CT exams and 1725 OPGs demonstrating feasibility of CT-based identification

Guideline-Based Recommendations

Diagnosis

• Utilize CT imaging of midface regions including lower and upper teeth rows, maxillary sinuses, cervical spine, and eye structures for identification
• Select CT slices manually from predefined anatomical regions for optimal feature extraction
• Exclude images with artifacts such as missing teeth, metal implants, or motion artifacts that hinder feature detection

Management

• Establish antemortem computer vision feature databases from CT and OPG images for matching unknown individuals
• Apply image preprocessing steps including color depth normalization, edge enhancement with modified Sobel filters, and noise reduction
• Use AKAZE algorithm parameters (Sobel gradient, averaging filter size, octaves, layers, diffusivity, descriptor type) optimized for CT images

Monitoring & Follow-up

• Evaluate identification success by matching highest scoring CV feature correspondences between unknown and database images
• Analyze top matching results to confirm identity or provide clues
• Monitor for image quality and artifact presence that may affect identification accuracy

Risks

• Potential failure of identification due to image artifacts such as metal implants or motion
• Requirement for manual slice selection may introduce variability
• Dependence on availability of corresponding antemortem images for database matching

Patient & Prescribing Data

Individuals undergoing cranial CT and OPG imaging, including trauma patients and deceased persons

Automated CV-based identification can assist in rapid and accurate personal identification without extensive manual segmentation or CNN training

Clinical Best Practices

• Ensure CT imaging protocols include midface visualization to capture key anatomical regions
• Maintain comprehensive antemortem CV feature databases for comparison
• Apply standardized image preprocessing and feature extraction parameters for consistency
• Exclude images with significant artifacts to improve identification reliability
• Use combined analysis of multiple anatomical regions to enhance identification accuracy

References

• 1-6: Challenges in unknown individual identification in emergency and forensic settings
• 7-10: Role of CT in emergency care and virtual autopsies
• 11-13: Previous CV-based personal identification methods using OPGs
• 14-18: Use of paranasal sinuses in CT for personal identification
• 19: AKAZE algorithm for feature extraction