Detection of Diabetic Retinopathy using Deep Learning and Transfer Learning Techniques with Oversampling to Address Imbalanced Dataset

Teodora RANĐELOVIĆ

Bakalářská práce

Detection of Diabetic Retinopathy using Deep Learning and Transfer Learning techniques with oversampling o address imbalanced dataset

Abstract:

The study aims to develop a system for detecting diabetic retinopathy using deep learning. In this study I have explored transfer learning with four distinct models and addressed the issue of an unbalanced dataset with oversampling. The final experiment achieved a significant improvement in accuracy and quadratic kappa score. The study highlights the potential of deep learning and the importance of addressing dataset imbalances for accurate results.

Abstract:

The study aims to develop a system for detecting diabetic retinopathy using deep learning. In this study I have explored transfer learning with four distinct models and addressed the issue of an unbalanced dataset with oversampling. The final experiment achieved a significant improvement in accuracy and quadratic kappa score. The study highlights the potential of deep learning and the importance of addressing dataset imbalances for accurate results.

Keywords

Diabetic Retinopathy, Deep learning, Transfer learning, Convolutional neural network, Image classification, medical imaging, diabetic macular edema, retinal fundus photographs, comparative analysis, oversampling, accuracy, quadratic kappa score

Jazyk práce: angličtina

Datum vytvoření / odevzdání či podání práce: 13. 5. 2023

Obhajoba závěrečné práce

• Vedoucí: Elisabeth Rumetshofer, MSc.

JIHOČESKÁ UNIVERZITA V ČESKÝCH BUDĚJOVICÍCH

Přírodovědecká fakulta

Bakalářský studijní program / obor:
Applied Informatics / Bioinformatics