AI in Eye Health: Model Comparison

This project aims to conduct a comparative evaluation of two advanced convolutional neural network (CNN) architectures, EfficientNet-V2 and ResNet, for the task of DR classification. The study utilizes two widely recognized datasets: APTOS 2019 Blindness Detection and IDRiD (Indian Diabetic Retinopathy Image Dataset), providing a diverse collection of retinal images annotated for DR severity levels. EfficientNet-V2, known for its balance between accuracy and computational efficiency, is compared against ResNet, a popular CNN model renowned for its strong performance in computer vision tasks. Both architectures are rigorously trained and optimized using techniques like transfer learning and data augmentation to ensure robust performance across the datasets. The models' performance is evaluated on their ability to classify retinal images into five DR severity categories: no DR, mild non-proliferative DR, moderate nonproliferative DR, severe non-proliferative DR, and proliferative DR. Comprehensive metrics are employed to assess the accuracy and reliability of the models' predictions. By comparing the performance of EfficientNet-V2 and ResNet on these diverse datasets, the study aims to provide insights into the suitability and potential of these architectures for DR classification tasks. The findings will contribute to the ongoing research efforts in developing accurate and reliable computer-aided diagnosis systems for diabetic retinopathy, ultimately supporting early detection and improving patient outcomes. **Target Audience:** - Machine Learning Experts - Individuals passionate about healthcare technology **Data for Model Training:** - The project will utilize publicly available datasets to train both EfficientNet-V2 and ResNet models. Accessibility to a variety of datasets ensures a broad and comprehensive training process, essential for achieving high accuracy and reliability in disease classification. **Development of User Interface:** - A straightforward, user-friendly interface will be developed. This UI is intended to display the classification results clearly and concisely, catering to healthcare professionals who may not have deep technical knowledge in machine learning but need to interpret the results effectively in their diagnostic process. **Ideal Skills and Experience for the Job:** - Proficiency in machine learning and deep learning, specifically with experience in EfficientNet-V2 and ResNet models. - Strong background in medical imaging analysis or a keen interest in healthcare applications of machine learning. - Experience in developing simple yet functional user interfaces, preferably with knowledge in UI/UX design principles tailored for medical applications. - Ability to work with publicly available datasets, including data pre-processing and augmentation techniques to improve model training. This project not only promises to advance the field of medical diagnostics through AI but also offers an opportunity to contribute to meaningful health outcomes for individuals affected by Diabetic Retinopathy. If you have the skills and the drive to help realize this vision, I look forward to your bid and embarking on this exciting journey together.