Scholarlink

Background <p>Diabetic retinopathy (DR) is a leading cause of adult blindness, stemming from prolonged diabetes that damages retinal blood vessels. Accurate early detection is essential but often challenging due to symptom overlap with other eye diseases.</p> Objectives <p>This study proposes an optimized deep learning technique to enhance the precision and speed of DR classification.</p> Methods <p>The proposed method, named a deep learning–based technique was used for DR classification using 1-dimensional complex-valued convolutional neural networks to improve early diagnosis and treatment outcomes (DRC-1-D CVCNN-IED), employs a 1-dimensional complex-valued convolutional neural network (1-D CVCNN) optimized by the parrot optimizer algorithm (POA). The input images are sourced from the Mendeley public database and undergo pre-processing using the constrained normalized subband adaptive filter (CNSAF) to clean, crop, and enhance image quality from the database. The pre-processed images are given to a 1-dimensional complex-valued convolutional neural network (1-D CVCNN) for detecting the eye disease and it classifies like choroidal neovascularization (CNV), diabetic macular oedema (DME), DRUSEN, and NORMAL. In general, there is no adaptation of optimization techniques using 1-D CVCNN to determine the optimal settings to provide precise classification. A POA is used to optimize the weight variables of the 1-D CVCNN method.</p> Results <p>The proposed technique is implemented using Python, and its effectiveness is assessed using various performance metrics, including precision, accuracy, F1 score, recall, and specificity. The DRC-1-D CVCNN-IED method shows a substantial improvement in CNV classification performance, achieving accuracy gains of 21.36%, 25.48%, and 15.27% compared to existing approaches such as OCTI-EDD-DCNN (optical coherence tomography image-based eye disease detection using deep CNN), DD-DED-DNN (diabetic eye disease detection using deep neural networks), and EDD-MSDL-CNN (multi-stage deep learning–based eye disease detection), respectively.</p> Conclusions <p>The DRC-1-D CVCNN-IED method demonstrates superior performance in diabetic eye disease classification, particularly in challenging CNV cases. The integration of POA with 1-D CVCNN contributes to precise and reliable DR detection, offering potential for clinical support systems.</p> Graphical abstract <p></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

A deep learning–based method for diabetic retinopathy classification using 1-dimensional complex-valued convolutional neural networks to improve early diagnosis and treatment outcomes

  • 期刊论文

  • 发表日期：

    2025年7月3日

查看全文
  • Subasree S.,
  • Sakthivel N. K.,
  • Baby Kalpana Y.,
  • Manikandan M.

摘要

Background

Diabetic retinopathy (DR) is a leading cause of adult blindness, stemming from prolonged diabetes that damages retinal blood vessels. Accurate early detection is essential but often challenging due to symptom overlap with other eye diseases.

Objectives

This study proposes an optimized deep learning technique to enhance the precision and speed of DR classification.

Methods

The proposed method, named a deep learning–based technique was used for DR classification using 1-dimensional complex-valued convolutional neural networks to improve early diagnosis and treatment outcomes (DRC-1-D CVCNN-IED), employs a 1-dimensional complex-valued convolutional neural network (1-D CVCNN) optimized by the parrot optimizer algorithm (POA). The input images are sourced from the Mendeley public database and undergo pre-processing using the constrained normalized subband adaptive filter (CNSAF) to clean, crop, and enhance image quality from the database. The pre-processed images are given to a 1-dimensional complex-valued convolutional neural network (1-D CVCNN) for detecting the eye disease and it classifies like choroidal neovascularization (CNV), diabetic macular oedema (DME), DRUSEN, and NORMAL. In general, there is no adaptation of optimization techniques using 1-D CVCNN to determine the optimal settings to provide precise classification. A POA is used to optimize the weight variables of the 1-D CVCNN method.

Results

The proposed technique is implemented using Python, and its effectiveness is assessed using various performance metrics, including precision, accuracy, F1 score, recall, and specificity. The DRC-1-D CVCNN-IED method shows a substantial improvement in CNV classification performance, achieving accuracy gains of 21.36%, 25.48%, and 15.27% compared to existing approaches such as OCTI-EDD-DCNN (optical coherence tomography image-based eye disease detection using deep CNN), DD-DED-DNN (diabetic eye disease detection using deep neural networks), and EDD-MSDL-CNN (multi-stage deep learning–based eye disease detection), respectively.

Conclusions

The DRC-1-D CVCNN-IED method demonstrates superior performance in diabetic eye disease classification, particularly in challenging CNV cases. The integration of POA with 1-D CVCNN contributes to precise and reliable DR detection, offering potential for clinical support systems.

Graphical abstract