التفاصيل البيبلوغرافية
| العنوان: |
Diabetic retinopathy diagnosis using deep learning ; Diagnóstico de la retinopatía diabética usando aprendizaje profundo |
| المؤلفون: |
Ariza López, Luis, Sanz Ramos, Álvaro |
| المساهمون: |
Sánchez Ruiz-Granados, Antonio Alejandro |
| سنة النشر: |
2023 |
| المجموعة: |
Universidad Complutense de Madrid (UCM): E-Prints Complutense |
| مصطلحات موضوعية: |
004(043.3), Deep learning, Convolutional neural networks, Classification, Computer vision, Diabetic retinopathy, Vision Transformers, Transfer learning, Model calibration, Model interpretability, Medical image analysis, Aprendizaje profundo, Redes neuronales convolucionales, Clasificación, Visión artificial, Retinopatía diabética, Aprendizaje transferido, Calibración del modelo, Interpretabilidad, Análisis de imágenes médicas, Informática (Informática), 33 Ciencias Tecnológicas |
| الوصف: |
Trabajo de Fin de Doble grado en Ingeniería Informática y Matemáticas, Facultad de Informática UCM, Departamento de Ingeniería de Software e Inteligencia Artificial, Curso 2022/2023. ; Diabetic retinopathy (DR) is one of the main complications of diabetes and the leading cause of new cases of blindness. Early detection is fundamental for a good prognosis, but diagnosis is a hard, expensive and time-consuming process. The need of automating methods for DR grading was recognized time ago, but most approaches in literature require vast computational power and have not been designed with interpretability in mind. We show that a model based in a convolutional neural network can achieve excellent performance at DR grading, comparing favorably to much larger models and achieving state-of-the-art results. Using transfer learning, we reduce to the minimum the computational requirements: the model can be trained in a few hours on domestic hardware. We use the hidden representation learned by the model to identify images diagnostically similar to a given one and explore the possibility of using this model in a clinical setting in a series of test carried on collaboration with a professional ophthalmologist. According to the criterion of the professional, the tool correctly identifies similar images and is a helpful assistance during diagnosis. Furthermore, we implement several interpretability tools to understand how the model makes predictions, address important concerns for clinical application (as calibration) and compare our approach to an alternative one using Vision Transformers under strict computational requirements. ; La retinopatía diabética es una de las principales complicaciones de la diabetes y la principal causa de nuevos casos de ceguera. Detectar esta enfermedad en sus primeras etapas es clave para un buen pronóstico, pero el diagnóstico es un proceso difícil, caro y que requiere tiempo. Para solventar estos inconvenientes se han desarrollado métodos que automatizan el diagnóstico, pero la mayor parte de ... |
| نوع الوثيقة: |
bachelor thesis |
| وصف الملف: |
application/pdf |
| اللغة: |
English |
| Relation: |
https://hdl.handle.net/20.500.14352/87739 |
| الاتاحة: |
https://hdl.handle.net/20.500.14352/87739 |
| Rights: |
Attribution-NonCommercial-NoDerivatives 4.0 International ; open access ; http://creativecommons.org/licenses/by-nc-nd/4.0/ |
| رقم الانضمام: |
edsbas.79EF0E39 |
| قاعدة البيانات: |
BASE |