Advancement and Enhancement of Artificial Intelligence in Diabetic Retinopathy Recognition: The Juxtaposition and Implications of Using Various Transfer Learning Tactics

作     者：Prasanth, S. Roshni Thanka, M. Bijolin Edwin, E. Veemaraj, Ebenezer 

作者机构：Division of Computer Science and Engineering Karunya Institute of Technology and Sciences Coimbatore India 

出 版 物：《SSRN》 

年 卷 期：2024年

核心收录：

主　　题：Convolutional neural networks 

摘      要：BackgroundDiabetic retinopathy, or DR, is still an important risk factor to progressive impairment of vision, especially in individuals who remain employed generation. Early detection is crucial to preventing blindness. Current deep learning techniques like convolutional neural network, deep neural network open up novel possibilities for crowdsourcing DR detection from ocular or retinal images. Identical problems like inadequate data and processing capacity can be solved with the aid of transfer learning. In order to improve DR advancement, this research assesses and compares the performance of several CNN infrastructures and transfer learning *** multiple techniques for preprocessing implemented in DR detection include color conversion, resizing, cropping, resizing, Ben Graham s preprocessing technique, and down-slicing. The various forfeiture-tuned models of transfer learning can be used, such as DenseNet, EfficientNet, GoogleNet, and others. A CNN model that has been developed specifically is used to evaluate these models performance on a simulated dataset. The model evaluation, or performance metrics, has been evaluated using the kappa value or QWK score *** the research findings, it s been adapted Transfer learning CNNs (EfficientNetB0, B1, B2, B3, B4, B5, B6, and DenseNet121) for the DR process range. In Pre-processing techniques, Ben Graham s preprocessing technique performed well for the proposed model. Additionally, healthcare image processing has been improved. EfficientNetB1 Iteration Approach achieved good precision. The performance of each model was also compared to the outcomes of the existing system, using QWK score as the performance *** findings provide invaluable insight into selecting appropriate transfer learning models based on computing capabilities and dataset features for deep learning findings. The automated DR finding technique is advanced by this analysis, helping diabetic retinopathy suffer