Automated longitudinal segmentation of lung and COVID-19 findings exploring self-supervised learning and polymorphic training

• Grant number: 22/02344-8
• Support Opportunities: Scholarships abroad - Research Internship - Doctorate
• Start date until: September 01, 2022
• End date until: February 28, 2023
• Field of knowledge: Engineering - Electrical Engineering
• Principal Investigator: Roberto de Alencar Lotufo
• Grantee: Diedre Santos do Carmo
• Supervisor: Joseph M. Reinhardt
• Host Institution: Faculdade de Engenharia Elétrica e de Computação (FEEC). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil
• Institution abroad: University of Iowa, United States
• Associated to the scholarship: 19/21964-4 - Lung diseases and COVID diagnosis from computed tomography images using explainable deep learning, BP.DR

Abstract

The use of deep learning for automated segmentation of the lung and COVID-19 findings using computed tomography (CT) images in the current COVID-19 pandemic is a very active research topic in medical imaging segmentation. Some automated segmentation methods have already been proposed with supervised learning over manual annotations. The lack of availability of manual annotations data paired with the possibility of overfitting to specific characteristics of the scarcely available data puts in check the reliability of those methods. We have already curated a large dataset of annotated CT data for supervised segmentation, and made avail-able a public segmentation method. For our next step in this project for a 6-month Research Internship Abroad (BEPE), we propose to attack two research directions identified through a recently concluded systematic review of the literature. With the intent to leverage more data that will be made available through this collaboration, firstly we will explore the use of self-supervised context learning, studying if a pre-training step with unannotated data could later improve the performance of supervised segmentation learning. Secondly, we will expand on a recent proposal of our collaborators, named polymorphic training, to explore if learning with different forms of annotations for COVID-19 findings is possible and if it leads to better results. Validation and ablation studies of those two hypothesis will beperformed through a longitudinal study of new annotated longitudinal COVID-19 patient data. (AU)

Scientific publications

(References retrieved automatically from Web of Science and SciELO through information on FAPESP grants and their corresponding numbers as mentioned in the publications by the authors)

CARMO, D. S.; TUDAS, R. A.; COMELLAS, A. P.; PEZZULO, A. A.; VILLACRESES, R. A.; CHO, J. L.; ZABNER, J.; RITTNER, L.; LOTUFO, R. A.; REINHARDT, J. M.; et al. Evidence of Post-infection Vaccination Correlation With the Severity of Opacities and Consolidations on the Lung Parenchyma of Post-acute SARS-CoV-2 Patients. American Journal of Respiratory and Critical Care Medicine, v. 207, p. 1-pg., 2023-05-01. (19/21964-4, 22/02344-8)