Loading...
Publication
Aspectos do rastreamento do glaucoma auxiliados por técnicas automatizadas em imagens com menor qualidade do disco óptico
| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 3.05 MB | Adobe PDF |
Authors
Abstract(s)
O glaucoma é uma neuropatia óptica cuja progressão pode levar a cegueira. Representa
a principal causa de perda visual de caráter irreversível em todo o mundo para homens
e mulheres. A detecção precoce através de programas de rastreamento feita por
especialistas é baseada nas características do nervo óptico, em biomarcadores
oftalmológicos (destacando-se a pressão ocular) e exames subsidiários, com destaque
ao campo visual e OCT. Após o reconhecimento dos casos é feito o tratamento com
finalidade de estacionar a progressão da doença e melhorar a qualidade de vida dos
pacientes. Contudo, estes programas têm limitações, principalmente em locais mais
distantes dos grandes centros de tratamento especializado, insuficiência de
equipamentos básicos e pessoal especializado para oferecer o rastreamento a toda a
população, faltam meios para locomoção a estes centros, desinformação e
desconhecimento da doença, além de características de progressão assintomática da
doença.
Esta tese aborda soluções inovadoras que podem contribuir para a automação do
rastreamento do glaucoma utilizando aparelhos portáteis e mais baratos, considerando
as necessidades reais dos clínicos durante o rastreamento.
Para isso foram realizadas revisões sistemáticas sobre os métodos e equipamentos para
apoio à triagem automática do glaucoma e os métodos de aprendizado profundo para
a segmentação e classificação aplicáveis. Também foi feito um levantamento de
questões médicas relativas à triagem do glaucoma e associá-las ao campo da inteligência
artificial, para dar mais sentido as metodologias automatizadas. Além disso, foi criado
um banco de dados privado, com vídeos e imagens de retina adquiridos por um
smartphone acoplado a lente de baixo custo para o rastreamento do glaucoma e
avaliado com métodos do estado da arte. Foram avaliados e analisados métodos de
detecção automática de glaucoma utilizando métodos de aprendizado profundo de
segmentação do disco e do copo óptico em banco de dados públicos de imagens de
retina. Finalmente, foram avaliadas técnicas de mosaico e de detecção da cabeça do
nervo óptico em imagens de baixa qualidade obtidas para pré-processamento de
imagens adquiridas por smartphones acoplados a lente de baixo custo.
Glaucoma is an optic neuropathy whose progression can lead to blindness. It represents the leading cause of irreversible visual loss worldwide for men and women. Early detection through screening programs carried out by specialists is based on the characteristics of the optic papilla, ophthalmic biomarkers (especially eye pressure), and subsidiary exams, emphasizing the visual field and optical coherence tomography (OCT). After recognizing the cases, the treatment is carried out to stop the progression of the disease and improve the quality of patients’ life. However, these screening programs have limitations, particularly in places further away from the sizeable, specialized treatment centers, due to the lack of essential equipment and technical personnel to offer screening to the entire population, due to the lack of means of transport to these centers, due to lack of information and lack of knowledge about the disease, considering the characteristics of asymptomatic progression of the disease. This thesis aims to develop innovative approaches to contribute to the automation of glaucoma screening using portable and cheaper devices, considering the real needs of clinicians during screening. For this, systematic reviews were carried out on the methods and equipment to support automatic glaucoma screening, and the applicable deep learning methods for segmentation and classification. A survey of medical issues related to glaucoma screening was carried out and associated with the field of artificial intelligence to make automated methodologies more effective. In addition, a private dataset was created, with videos and retina images acquired using a low-cost lens-coupled cell phone, for glaucoma screening and evaluated with state-of-the-art methods. Methods of automatic detection of glaucoma using deep learning methods of segmentation of the disc and optic cup were evaluated and analyzed in a public database of retinal images. In the case of deep learning classification methods, these were evaluated in public databases of retina images and in a private database with low-cost images. Finally, mosaic and object detection techniques were evaluated in low-quality images obtained for pre-processing images acquired by cell phones coupled with low-cost lenses.
Glaucoma is an optic neuropathy whose progression can lead to blindness. It represents the leading cause of irreversible visual loss worldwide for men and women. Early detection through screening programs carried out by specialists is based on the characteristics of the optic papilla, ophthalmic biomarkers (especially eye pressure), and subsidiary exams, emphasizing the visual field and optical coherence tomography (OCT). After recognizing the cases, the treatment is carried out to stop the progression of the disease and improve the quality of patients’ life. However, these screening programs have limitations, particularly in places further away from the sizeable, specialized treatment centers, due to the lack of essential equipment and technical personnel to offer screening to the entire population, due to the lack of means of transport to these centers, due to lack of information and lack of knowledge about the disease, considering the characteristics of asymptomatic progression of the disease. This thesis aims to develop innovative approaches to contribute to the automation of glaucoma screening using portable and cheaper devices, considering the real needs of clinicians during screening. For this, systematic reviews were carried out on the methods and equipment to support automatic glaucoma screening, and the applicable deep learning methods for segmentation and classification. A survey of medical issues related to glaucoma screening was carried out and associated with the field of artificial intelligence to make automated methodologies more effective. In addition, a private dataset was created, with videos and retina images acquired using a low-cost lens-coupled cell phone, for glaucoma screening and evaluated with state-of-the-art methods. Methods of automatic detection of glaucoma using deep learning methods of segmentation of the disc and optic cup were evaluated and analyzed in a public database of retinal images. In the case of deep learning classification methods, these were evaluated in public databases of retina images and in a private database with low-cost images. Finally, mosaic and object detection techniques were evaluated in low-quality images obtained for pre-processing images acquired by cell phones coupled with low-cost lenses.
Description
Keywords
Triagem do glaucoma Aprendizado profundo Retinal images Mobile devices Segmentation Glaucoma classification Disco óptico Redes Neurais Convolucionais (CNN) Cup CAD Glaucoma screening Deep learning Retinal images Mobile devices Optical disc
Citation
Camara, José Carlos Raposo da - Aspectos do rastreamento do glaucoma auxiliados por técnicas automatizadas em imagens com menor qualidade do disco óptico [Em linha]. [S.l.]: [s.n.], 2023. 189 p.