Camilla Alvarez will present a Poster at the conference.
Machine Learning on Retina Images for Diagnostic Decision Support
Maria Camila Alvarez Triviño1, Jesús Alfonso López Sotelo 1, Jérémie Despraz 2,3, and Carlos Peña-Reyes 2,3
1 Universidad Autónoma de Occidente, Colombia
2 School of Business and Engineering Vaud (HEIG-VD), University of Applied Sciences of Western Switzerland (HES-SO), Yverdon-Les-Bains, Switzerland
3 SIB Swiss Institute of Bioinformatics, Lausanne, Switzerland
Abstract
In this project we developed a deep learning system applied to human retina images for medical diagnostic decision support. The retina images were provided by EyePACS. These images were used in a Kaggle contest, whose purpose was identifying diabetic retinopathy signs through an automatic detection system. Using as inspiration one of the solutions proposed in the contest, we implemented a model to detect diabetic retinopathy from input retinal images. A preprocessing was applied to the images and then, they were used as input to a deep convolutional neural network (CNN). The CNN performed a feature extraction process followed by a classification stage, which allowed the system to differentiate between healthy and sick patients (using five categories). Finally, we created a system capable of identifying this pathology with an agreement rate with respect to the medical expert’s labels of 76,73% for test data.
In this project we developed a deep learning system applied to human retina images for medical diagnostic decision support. The retina images were provided by EyePACS. These images were used in a Kaggle contest, whose purpose was identifying diabetic retinopathy signs through an automatic detection system. Using as inspiration one of the solutions proposed in the contest, we implemented a model to detect diabetic retinopathy from input retinal images. A preprocessing was applied to the images and then, they were used as input to a deep convolutional neural network (CNN). The CNN performed a feature extraction process followed by a classification stage, which allowed the system to differentiate between healthy and sick patients (using five categories). Finally, we created a system capable of identifying this pathology with an agreement rate with respect to the medical expert’s labels of 76,73% for test data.