Fully automated radiologic identification focusing on the sternal bone

Area of Science:

• Forensic Science
• Radiology
• Medical Imaging

Background:

• Accurate identification of unknown deceased is critical in forensic investigations.
• Current morphologic identification methods often lack standardization and statistical validation.
• Automated radiologic identification methods are needed to overcome existing challenges.

Purpose of the Study:

• To develop a fully automated radiologic identification (autoRADid) method based on sternal bone analysis.
• To overcome limitations of current manual and subjective identification techniques.

Main Methods:

• A custom Python pipeline was developed for automated registration of ante mortem (AM) and post mortem (PM) chest CT scans.
• A two-step registration method was employed to align AM and PM data.
• Image similarity measures (Jaccard Coefficient, Dice Coefficient, Mutual Information) were used to evaluate identification success.

Main Results:

• The autoRADid method correctly identified 38 out of 42 cases, achieving an accuracy of 91.2%.
• Successful identification was based on comparing AM and PM sternal bone data using CT scans.
• Four cases were unsuccessfully matched due to surgical interventions or poor CT scan quality.

Conclusions:

• The autoRADid method presents a promising, fully automated tool for reliable and efficient identification of unknown deceased individuals.
• The developed pipeline is open source and available for future forensic applications.
• This automated approach enhances the objectivity and statistical evidence in forensic identification.