Unsupervised learning for fundus image registration via convolutional neural network and optimial mass transportation

• Grant number: 19/26288-7
• Support Opportunities: Scholarships in Brazil - Master
• Start date: June 01, 2020
• End date: February 28, 2022
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Wallace Correa de Oliveira Casaca
• Grantee: Giovana Augusta Benvenuto
• Host Institution: Faculdade de Ciências e Tecnologia (FCT). Universidade Estadual Paulista (UNESP). Campus de Presidente Prudente. Presidente Prudente , SP, Brazil
• Associated research grant: 13/07375-0 - CeMEAI - Center for Mathematical Sciences Applied to Industry, AP.CEPID

Abstract

The use of imaging exams is very common in the ophthalmology field. In this medical branch, fundus images acquired at different moments or from different equipments are used to aid medical diagnoses as well as to track the progress of diseases. Such a problem is known in the medical imaging literature as image registration, which has been experiencing important advances due to the systematic use of Machine Learning (ML). Despite the recent progress made in image registration, there is no consensus regarding the existence of a definitive technique that simultaneously reaches an acceptable computational cost, high accuracy and elevated operability in practical contexts of usage. In addition, existing techniques either rely on ground-truth data to be effective, i.e., they are supervised or do not meet the aforementioned requirements. In addition, the image registration literature exploited under the ML perspective is relatively new, especially the unsupervised retinal registration one. Therefore, this research project aims at addressing the problem of retinal image registration, by introducing a new fully unsupervised neural network which will be capable of dealing with both the matching between image structures as well as the registration task itself. The proposed ML pipeline combines Optimal Transport theory, Convolutional Neural Network and a new overlap metric for retinal images into a concise and unified framework for unsupervised image registration. Moreover, the proposed approach will be evaluated in a real medical context by two experienced ophthalmologists in the treatment and registration of fundus images, thus allowing for assessing the impact of our registration tool in real medical applications. (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)

BENVENUTO, GIOVANA A.; COLNAGO, MARILAINE; DIAS, MAURICIO A.; NEGRI, ROGERIO G.; SILVA, ERIVALDO A.; CASACA, WALLACE. A Fully Unsupervised Deep Learning Framework for Non-Rigid Fundus Image Registration. BIOENGINEERING-BASEL, v. 9, n. 8, p. 17-pg., 2022-08-01. (13/07375-0, 21/01305-6, 19/26288-7, 21/03328-3)

BENVENUTO, GIOVANA AUGUSTA; COLNAGO, MARILAINE; CASACA, WALLACE; IEEE. UNSUPERVISED DEEP LEARNING NETWORK FOR DEFORMABLE FUNDUS IMAGE REGISTRATION. 2022 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP), v. N/A, p. 5-pg., 2022-01-01. (13/07375-0, 19/26288-7, 21/03328-3)

Academic Publications

(References retrieved automatically from State of São Paulo Research Institutions)

BENVENUTO, Giovana Augusta. Retinal images registration via unsupervised deep learning. 2022. Master's Dissertation - Universidade Estadual Paulista (Unesp). Faculdade de Ciências e Tecnologia. Presidente Prudente Presidente Prudente.