Scholarlink

<p>Among individuals with diabetes, Diabetic Retinopathy (DR) stands out as a primary cause of vision loss and potential blindness. Many of the already existing diagnostic methods are complex and require an expert eye to give any meaningful interpretation, thus making a case for the application of the latest technology in fast-tracking disease detection. These models have precedence to improve diagnostic accuracy and speed over traditional processes. Motivated by these deep learning opportunities, this paper presents the DiabEyeNet model dedicated to DR detection and classification. The DiabEyeNet model integrates the Residual Inception Multimodal Fusion Network (RIM-FuseNet) with the Crested Porcupine Chaotic Levy Optimization (CPCLO) to mitigate the problems associated with DR identification. The RIM-FuseNet part takes inspiration from residual learning, which guides the model to extract and further distill more precise features for increasing the model’s response to different DR stages. CPCLO optimization technique adjusts the model parameters and improves the performance indicators of the model and computational efficiency. The nature of the DiabEyeNet model, as a framework, reveals reduced computational time and reduced error, showing efficiency and practicability for this model. The proposed DiabEyeNet model is demonstrated to be superior in diabetic retinopathy diagnosis with 99.2% accuracy, 99% sensitivity, 98.7% specificity, and AUC (Area under the Curve) of 0.996, which reflects its strong discriminative ability for all DR severity grades. DiabEyeNet’s computational speed is also acceptable, at a processing time of only 7.3&#xa0;s on average per individual fundus image, which renders it feasible for real-time or near-real-time clinical screening use.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

DiabEyeNet: Integrating Residual Inception Networks and Chaotic Optimization for Robust Diabetic Retinopathy Detection

  • 期刊论文

  • 发表日期：

    2025年8月22日

查看全文
  • S. Chandravadhana,
  • V. Anusuya,
  • D. Kirubha,
  • P. Archana,
  • Pyla Ravikiran

摘要

Among individuals with diabetes, Diabetic Retinopathy (DR) stands out as a primary cause of vision loss and potential blindness. Many of the already existing diagnostic methods are complex and require an expert eye to give any meaningful interpretation, thus making a case for the application of the latest technology in fast-tracking disease detection. These models have precedence to improve diagnostic accuracy and speed over traditional processes. Motivated by these deep learning opportunities, this paper presents the DiabEyeNet model dedicated to DR detection and classification. The DiabEyeNet model integrates the Residual Inception Multimodal Fusion Network (RIM-FuseNet) with the Crested Porcupine Chaotic Levy Optimization (CPCLO) to mitigate the problems associated with DR identification. The RIM-FuseNet part takes inspiration from residual learning, which guides the model to extract and further distill more precise features for increasing the model’s response to different DR stages. CPCLO optimization technique adjusts the model parameters and improves the performance indicators of the model and computational efficiency. The nature of the DiabEyeNet model, as a framework, reveals reduced computational time and reduced error, showing efficiency and practicability for this model. The proposed DiabEyeNet model is demonstrated to be superior in diabetic retinopathy diagnosis with 99.2% accuracy, 99% sensitivity, 98.7% specificity, and AUC (Area under the Curve) of 0.996, which reflects its strong discriminative ability for all DR severity grades. DiabEyeNet’s computational speed is also acceptable, at a processing time of only 7.3 s on average per individual fundus image, which renders it feasible for real-time or near-real-time clinical screening use.