Detalhes bibliográficos
| Ano de defesa: |
2021 |
| Autor(a) principal: |
SOUSA, Jefferson Alves de |
| Orientador(a): |
PAIVA, Anselmo Cardoso de |
| Banca de defesa: |
PAIVA, Anselmo Cardoso de,
SILVA, Aristófanes Corrêa,
BRAZ JÚNIOR, . Geraldo,
ALMEIDA, João Dallyson Sousa de,
AIRES, Kelson Rômulo Teixeira |
| Tipo de documento: |
Tese
|
| Tipo de acesso: |
Acesso aberto |
| Idioma: |
por |
| Instituição de defesa: |
Universidade Federal do Maranhão
|
| Programa de Pós-Graduação: |
PROGRAMA DE PÓS-GRADUAÇÃO EM ENGENHARIA DE ELETRICIDADE/CCET
|
| Departamento: |
DEPARTAMENTO DE ENGENHARIA DA ELETRICIDADE/CCET
|
| País: |
Brasil
|
| Palavras-chave em Português: |
|
| Palavras-chave em Inglês: |
|
| Área do conhecimento CNPq: |
|
| Link de acesso: |
https://tedebc.ufma.br/jspui/handle/tede/3411
|
Resumo: |
This thesis presents two fully automatic methods for diagnosing age-related macular degeneration (AMD) and diabetic macular edema (DME) through the use of image processing and pattern recognition techniques applied to optical coherence tomography (OCT) examinations. Age-related macular degeneration and diabetic macular edema are eye diseases that can cause visual impairment. Macular degeneration mainly affects older adults over 50 years of age, whereas macular edema is a consequence of diabetic retinopathy, affecting people with diabetes. Optical coherence tomography is one of the main tests for detecting and monitoring eye diseases. The evaluation of alterations caused by AMD and DME from OCT images is done by evaluating successive sectional cuts in search of morphological alterations. The use of CAD (Computer-Aided Detection) and CADx (Computer-Aided Diagnosis) systems has contributed to increasing the chances of correct detection and diagnosis, helping specialists make decisions about the treatment of these illnesses. Thus, a method was developed that uses segmentation of the edges of the retinal layers in OCT images made with two deep neural networks, U-Net and DexiNed, to delimit the edges. The classification uses a deep residual neural network. The second method for the automatic detection of AMD and DME uses a Capsule Network architecture with layers based on the Local Binary Pattern. The proposed segmentation method proved to be promising, reaching an overall mean absolute error of 0.49 pixels for the inner limiting membrane (ILM), 0.57 for the retinal pigment epithelium (RPE), and 0.66 for Bruch’s membrane (BM). The classification methods reached, respectively, an accuracy of 99% and 98% among multiple classes (Drusen, Choroidal Neovascularization, DME, and Normal). |
