Unveiling Identity Through Anatomy: Person Verification Using Vision Transformers on Chest X-Rays Radiographs

Context: The prospective utilization of medical imaging data for reliable individual identification and authentication has garnered significant interest in both security and healthcare sectors. This importance is particularly amplified during disaster scenarios, where conventional means of human verification become ineffective. In these challenging conditions, Chest X-rays serve as an essential resource by capturing unique anatomical details of the rib cage, lungs, and heart features that persist as reliable verification even when the body is compromised. Objective: We propose the creation of an innovative verification system for image retrieval, specifically designed to enhance person verification using chest X-ray images. Method: The system integrates a deep learning paradigm, leveraging Triplet network architecture; while uniquely use cosine similarity as a metric to assess similarity and dissimilarity between image features. Such an approach enables a more nuanced and robust feature comparison, leading to improved retrieval accuracy and verification performance. Building upon this premise, we propose the creation of an innovative verification system for image retrieval. Notably, our framework employs a state-of-the-art Vision Transformer (ViT) as the backbone for the Triplet network. Results: The ViT backbone offers robust capabilities in extracting and contextualizing features, thereby enhancing the discriminative power of the triplet architecture and ensuring improved retrieval accuracy. This novel integration not only expands the existing toolkit for medical image analysis but also reinforces the reliability of identity verification systems. Conclusions: The dual use of geometric and angular similarity measures, coupled with the advanced feature extraction of the ViT, offers a precise and dependable solution, particularly in high-stakes scenarios such as emergencies and security-critical applications.