Domain Invariant Detection of Medical Devices in Plain Chest X-ray Images

• Grant number: 24/00789-8
• Support Opportunities: Scholarships in Brazil - Doctorate
• Start date: March 01, 2024
• End date: February 28, 2025
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: João Paulo Papa
• Grantee: João Renato Ribeiro Manesco
• Host Institution: Faculdade de Ciências (FC). Universidade Estadual Paulista (UNESP). Campus de Bauru. Bauru , SP, Brazil
• Associated research grant: 13/07375-0 - CeMEAI - Center for Mathematical Sciences Applied to Industry, AP.CEPID

Abstract

Chest X-ray images play a crucial role in modern medicine, being an accessible and easy-to-use tool for radiologists to identify diseases and assess the proper positioning of medical devices. Ensuring the correct positioning of these devices is vital for the patient's health, especially in high-risk cases where X-ray examinations are frequent. With the advance of deep learning techniques to aid diagnosis, there is a need to accurately detect the positioning of medical devices in X-ray images. However, off-the-shelf methods often have many limitations, such as high cost and differences in acquisition protocols between institutions, which affects effectiveness. In this project, we intend to use data from different domains when training the models, in order to achieve robust out-of-distribution generalization capabilities. In addition, we intend to evaluate the clinical effectiveness of our approach through inter-institutional cross-validation. To date, we have found no methods in the literature that objectively describe domain shift in X-ray images and that offer solutions to solve the problem of medical device detection in this context. We hope to be able to develop a method, based on neural networks, that can solve this problem and address all the points discussed.