Publicação
Deep Learning Methods for Diabetic Eye Disease Screening and Smartphone-based Applications
| Resumo: | Vision is an essential part of daily life, for many this can be disrupted by several diseases like diabetes and cause severe eye complications or even blindness. With frequent eye screening early detection and prevention of blindness can be possible, however, there is still a global discrepancy in the availability and accessibility of eye care. In this Master’s Thesis, the possibility of providing affordable and efficient automatic diabetic eye disease screening using low-cost lenses, smartphones, and AI methods with the current technology is investigated and applications in the medical practice are discussed. A comprehensive review was performed to analyze the current state of the art AI methods. A market review about the products related to mobile ophthalmology was conveyed for affordability analysis. As part of the collaborative work, a systematic review was also conducted on AI applications for Inherited Retinal Disorders. On the other hand, testing was carried out to assess the efficiency of the chosen AI methods and both the dataset and base architecture impact on the AI algorithm were assessed by statistical analysis methods. Results suggest that multi-disease approaches perform slightly better than disease-specific ones where DR detection has higher values, and EfficientNet architectures achieve higher accuracy scores. All in all, using DL algorithms, automatic diabetic eye disease screening in mobile settings is achievable in the medical practice and can have huge impacts in the rural areas and less developed regions around the world. |
|---|---|
| Autores principais: | Esengönül, Meltem |
| Assunto: | Artificial Intelligence Deep Learning Diabetic Eye Diseases Ophthalmology Smartphone-based Technologies |
| Ano: | 2022 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso restrito |
| Instituição associada: | Universidade de Trás-os-Montes e Alto Douro |
| Idioma: | inglês |
| Origem: | Repositório da UTAD |
| Resumo: | Vision is an essential part of daily life, for many this can be disrupted by several diseases like diabetes and cause severe eye complications or even blindness. With frequent eye screening early detection and prevention of blindness can be possible, however, there is still a global discrepancy in the availability and accessibility of eye care. In this Master’s Thesis, the possibility of providing affordable and efficient automatic diabetic eye disease screening using low-cost lenses, smartphones, and AI methods with the current technology is investigated and applications in the medical practice are discussed. A comprehensive review was performed to analyze the current state of the art AI methods. A market review about the products related to mobile ophthalmology was conveyed for affordability analysis. As part of the collaborative work, a systematic review was also conducted on AI applications for Inherited Retinal Disorders. On the other hand, testing was carried out to assess the efficiency of the chosen AI methods and both the dataset and base architecture impact on the AI algorithm were assessed by statistical analysis methods. Results suggest that multi-disease approaches perform slightly better than disease-specific ones where DR detection has higher values, and EfficientNet architectures achieve higher accuracy scores. All in all, using DL algorithms, automatic diabetic eye disease screening in mobile settings is achievable in the medical practice and can have huge impacts in the rural areas and less developed regions around the world. |
|---|