Full text
| |
| Author(s): |
Giovana Augusta Benvenuto
Total Authors: 1
|
| Document type: | Master's Dissertation |
| Press: | Presidente Prudente. 2022-04-04. |
| Institution: | Universidade Estadual Paulista (Unesp). Faculdade de Ciências e Tecnologia. Presidente Prudente |
| Defense date: | 2022-03-03 |
| Advisor: | Wallace Correa de Oliveira Casaca |
| Abstract | |
The image registration problem consists of finding a geometric transformation which aligns a given set of images. Such a problem is important in the medical context, where it the use of exams taken from images is a common task in order to support diagnoses and monitor the progress of diseases, especially in specialties such as ophthalmology. On this field of study, retinal (fundus) images are used under different circumstances, as they have to be compared with other particular images acquired at different times or even by different devices, which makes this task difficult to be carried out manually. Advances in technology have allowed a growth in the registration literature, which employs Artificial intelligence techniques to solve such a problem. Despite recent progress in this area, there is no consensus on an ideal methodology for the systematic daily use of a registration tool by ophthalmic professionals. Therefore, in this work, we introduce a framework for registering fundus images by combining a Convolutional Neural Network with a Spatial Transformation module. We also take a similarity metric to calculate the loss of network in order to have a fully unsupervised pipeline, thus discarding the use of ground-truth data when carrying out the registration task. From an extensive battery of experimental tests, we conclude that our trained model was capable of dealing with different categories of fundus images while surpassing recent optimization and Deep Learning-based methods from the registration literature. (AU) | |
| FAPESP's process: | 19/26288-7 - Unsupervised learning for fundus image registration via convolutional neural network and optimial mass transportation |
| Grantee: | Giovana Augusta Benvenuto |
| Support Opportunities: | Scholarships in Brazil - Master |