Use this link to cite:
http://hdl.handle.net/2183/31199 Registro de imagen oftalmológica utilizando Deep Learning
Loading...
Identifiers
Publication date
Authors
Lin, Xin
Advisors
Other responsabilities
Enxeñaría informática, Grao en
Journal Title
Bibliographic citation
Type of academic work
Academic degree
Abstract
[Resumen] El registro de imagen es una etapa fundamental en el análisis de imagen médica para
realizar diagnósticos o seguimiento de enfermedades en los pacientes. Los métodos clásicos
de registro, requieren tiempo y, a menudo, tienen componentes específicas para cada tipo de
imagen. Recientemente se han propuesto técnicas basadas en Deep Learning (DL) para eliminar
la necesidad de diseñar extractores de características específicos para cada modalidad
de imagen, además con estos modelos el tiempo de registro se reduce drásticamente. En este
Trabajo Fin de Grado, se evalúa el funcionamiento de un modelo basado en DL sobre imágenes
oftalmológicas. Se consiguen resultados que indican que la utilización de estos modelos
sobre las mencionadas imágenes es posible, a través de un entrenamiento supervisado. Se experimentan
con diferentes modalidades de imagen, aunque de forma independiente (registro
monomodal), en los que se obtienen resultados similares, por lo sugiere que se puede utilizar
una misma arquitectura para diferentes tipos de imágenes.
[Abstract] Image registration is a fundamental step in medical image analysis for diagnosis or disease monitoring in a patient. Classical methods for image registration are of high time complexity and require specifically crafted components for different image types. Recently, techniques based on Deep Learning (DL) were proposed to solve the need of designing specific feature extractors for each image modality, also these models can perform the registration with lower execution times. In this work, these models are evaluated using ophthalmic images of different modalities. The results obtained indicate that these models can be used for training ophthalmic images for image registration in a supervised manner. Experimentation is performed using different modalities, although trained as monomodal registration tasks, where the similar results suggests that different type of images can be trained using the same network architecture.
[Abstract] Image registration is a fundamental step in medical image analysis for diagnosis or disease monitoring in a patient. Classical methods for image registration are of high time complexity and require specifically crafted components for different image types. Recently, techniques based on Deep Learning (DL) were proposed to solve the need of designing specific feature extractors for each image modality, also these models can perform the registration with lower execution times. In this work, these models are evaluated using ophthalmic images of different modalities. The results obtained indicate that these models can be used for training ophthalmic images for image registration in a supervised manner. Experimentation is performed using different modalities, although trained as monomodal registration tasks, where the similar results suggests that different type of images can be trained using the same network architecture.
Description
Editor version
Rights
Atribución-NoComercial 3.0 España








