Momentos geométricos y machine learning aplicados al estudio de datos oftalmológicos
Loading...
Download
Official URL
Full text at PDC
Publication date
2020
Authors
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citation
Abstract
El glaucoma es la primera causa de ceguera irreversible en el ámbito mundial. En su desarrollo, aumenta la excavación presente en la papila óptica de la retina. Esta depresión puede ser examinada mediante la tomografía de coherencia óptica y parametrizada como una superficie regular mediante la utilización de momentos geométricos invariantes a transformaciones de semejanza. Estas entidades matemáticas se pueden agrupar para formar un conjunto de datos de pacientes glaucomatosos y personas sanas que puede ser estudiado mediante técnicas estadísticas básicas y de aprendizaje automático. En este trabajo, se analiza dicho conjunto para evaluar el potencial de los momentos geométricos y de los métodos de machine learning en la investigación con datos oftalmológicos. Se estudian propiedades y características de los invariantes que permiten mejorar su entendimiento y aportar información sobre la enfermedad. Algunas de las cualidades que se examinan son la consistencia, la dispersión, la correlación entre los momentos, cómo se estructuran en el espacio, la relevancia de cada invariante en el diagnóstico del glaucoma y las muestras del entorno de los distintos elementos del conjunto en el espacio. Para abordar estas cuestiones, se hará uso de técnicas estadísticas como los estadísticos descriptivos más comunes, el coeficiente de correlación de Pearson y el error cuadrático medio; y de métodos de machine learning como los algoritmos de clustering, el análisis de componentes principales y los árboles de decisión. De esta forma, se lleva a cabo una primera prueba de concepto, sobre los datos en bruto, con la que se investigan las distintas cuestiones que pueden abordarse y con la que se plantean nuevos temas y preguntas, de manera que se alcancen conclusiones que aporten información sobre esta neuropatía óptica desde el campo de las matemáticas a la rama de la oftalmología.
Glaucoma is the leading cause of irreversible blindness worldwide. In its development, it increases the excavation present in the optic disc of the retina. This depression can be examined by optical coherence tomography and parameterised as a regular surface by using three-dimensional surface moments invariant under similarity transformations. Thanks to these mathematical entities, we have a data set of glaucomatous patients and healthy people that can be studied by means of basic statistical techniques and machine learning methods. In this work, this set is analysed to evaluate the potential of surface moments and machine learning methods in ophthalmological data research. Properties and characteristics of the invariants are studied to improve their understanding and provide information about the disease. Some of the qualities examined are consistency, dispersion, correlation between the moments, how they are structured in space, the relevance of each invariant in the diagnosis of glaucoma and the environmental samples of the different elements of the set in space. To address these issues, we will use statistical techniques such as the most common descriptive statistics, the Pearson's correlation coeficient and the mean square error; and machine learning methods such as clustering algorithms, principal component analysis and decision trees. In this way, a first proof of concept is carried out, on this raw data, with which the different issues that can be addressed are researched and with which new topics and questions are raised, so that conclusions are reached that provide information on this optical neuropathy from the field of mathematics to the branch of ophthalmology.
Glaucoma is the leading cause of irreversible blindness worldwide. In its development, it increases the excavation present in the optic disc of the retina. This depression can be examined by optical coherence tomography and parameterised as a regular surface by using three-dimensional surface moments invariant under similarity transformations. Thanks to these mathematical entities, we have a data set of glaucomatous patients and healthy people that can be studied by means of basic statistical techniques and machine learning methods. In this work, this set is analysed to evaluate the potential of surface moments and machine learning methods in ophthalmological data research. Properties and characteristics of the invariants are studied to improve their understanding and provide information about the disease. Some of the qualities examined are consistency, dispersion, correlation between the moments, how they are structured in space, the relevance of each invariant in the diagnosis of glaucoma and the environmental samples of the different elements of the set in space. To address these issues, we will use statistical techniques such as the most common descriptive statistics, the Pearson's correlation coeficient and the mean square error; and machine learning methods such as clustering algorithms, principal component analysis and decision trees. In this way, a first proof of concept is carried out, on this raw data, with which the different issues that can be addressed are researched and with which new topics and questions are raised, so that conclusions are reached that provide information on this optical neuropathy from the field of mathematics to the branch of ophthalmology.