The robustness of a Neural Network: Analyzing the effect of intrinsic biometric data on accuracy results.

• Grant number: 25/19762-5
• Support Opportunities: Scholarships abroad - Research Internship - Scientific Initiation
• Start date: December 18, 2025
• End date: February 19, 2026
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: João Paulo Papa
• Grantee: Maurício José Grapéggia Zanella
• Supervisor: Bruno Hochhegger
• Host Institution: Faculdade de Ciências (FC). Universidade Estadual Paulista (UNESP). Campus de Bauru. Bauru , SP, Brazil
• Institution abroad: University of Florida, Gainesville (UF), United States
• Associated to the scholarship: 24/01336-7 - Validation of Deep learning-based patient re-identification and use of adversarial techniques, BP.IC

Abstract

This project aims to explore flaws in anonymity techniques via deep learning algorithms and display the current issues with open-access datasets. Radiography is an essential technology in the medical field, and its effects on the landscape of diagnosis of different problems cannot be understated. The fast advancement of machine learning techniques, coupled with this intrinsic technology, has led to the development of computer-assisted diagnosis. With this comes a necessity for more robust, complete, and, most of all, anonymous datasets. Unfortunately, this advancement of machine learning also creates breaches that can be taken advantage of because of the hidden biometric data inside X-ray images. The student's main project uses deep siamese networks to verify their robustness for person identification using X-ray data in a Brazilian dataset. The current proposal aims to investigate the proposed methodology on a more significant dataset, curated by Prof. Bruno from the University of Florida. Besides access to the dataset, this BEPE proposal will provide the student with international experience and knowledge from a global and renowned radiologist. We aim to understand better how to prevent biometric identification from X-ray images in this project.