همایش ملی بیوانفورماتیک ایران

صفحه اصلی / 4th international edition and 13th Iranian Conference on Bioinformatics

Diagnosis of Diabetic Retinopathy with Fuzzy Technique and Deep Learning

نویسندگان :

Samaneh Noroozi¹ Sajad Haghzad Klidbary²

1- دانشگاه زنجان 2- دانشگاه زنجان

کلمات کلیدی :

Diabetic Retinopathy،Deep Learning (DL)،Fuzzy Logic،Fundus،Image Processing

چکیده :

Diabetic retinopathy is a growing eye disease and one of the most common complications of diabetes worldwide. If not diagnosed in time, it can lead to significant vision problems and eventually blindness. Early detection, before specific symptoms occur, is crucial for preserving vision. However, manual detection of retinal fundus images by ophthalmologists is both time-consuming and costly. This paper introduces a hybrid approach that uses deep learning and fuzzy techniques to classify retinal fundus images into five categories: healthy, mild retinopathy, moderate retinopathy, severe retinopathy, and proliferative retinopathy. Deep learning methods are known for their high accuracy, speed, and ability to automatically extract features from images. However, their performance is highly dependent on the quality and balance of the dataset. The dataset used in this study is Aptos2019 from Kaggle, which contains 5590 images divided into training and test sets, which shows significant imbalance among disease categories. In this study, fundus images are first processed using deep learning techniques to extract image features. The accuracy and extracted features are compared in a simple manual architecture and several popular deep learning frameworks. Then, the most effective features are fed to a fuzzy system to classify the disease into five categories. Fuzzy techniques, with their flexibility and similarity to human reasoning, allow us to calculate the probability percentage for each stage of the disease based on the extracted features. The rules governing the fuzzy system are interpretable by physicians and can be modified using their diagnosis, resulting in outputs that accurately reflect human decision-making. In this research, we attempt to develop a robust and comprehensive system that can be applied to any medical data and achieve desired results. The aim of this study is to combine the strengths of deep learning and fuzzy logic to inspire the development of innovative methods in the field of medical diagnosis and data analysis.