Channel and Spatial Attention in Chest X-Ray Radiographs: Advancing Person Identification and Verification with...

Area of Science:

• Medical Imaging
• Biometrics
• Artificial Intelligence

Background:

• Traditional biometrics (facial, fingerprint, iris, DNA) have limitations.
• Chest X-rays capture unique anatomical details (skeletal structure, organs).
• Chest X-rays are valuable for identification when bodies are damaged or disfigured, particularly in forensics.

Purpose of the Study:

• To introduce a novel deep learning approach for chest X-ray-based identity verification.
• To develop an effective feature embedding method for large-scale chest X-ray image verification.
• To enhance person identification and verification using chest X-ray categorization.

Main Methods:

• Developed a self-residual attention network (SRAN) incorporating self-channel and self-spatial attention modules.
• Implemented SRAN within a ResNet50 framework using self-residual attention blocks (SRAB).
• Trained a Siamese network with triplet loss for improved feature embedding in identity identification and verification.

Main Results:

• Achieved notable accuracy improvements in identity verification and identification using chest X-ray images.
• Demonstrated the method's effectiveness in capturing detailed individual anatomical characteristics.
• Validated chest X-rays as a viable biometric tool, even for damaged or disfigured individuals.

Conclusions:

• The SRAN method offers a promising solution for biometric identification using chest X-rays.
• This approach provides accurate and reliable identity verification, especially when traditional biometrics fail (e.g., postmortem, forensic cases).
• The methodology has transformative potential for biometric security and healthcare applications.