Scholarlink

Background <p>Diabetic retinopathy (DR) is a diabetes-induced retinal disease characterized by damage to the blood vessels of the retina. If left undiagnosed, DR can lead to vision impairment or blindness. The asymptomatic nature of the disease, often referred to as a “silent disease,” makes early detection critical. Traditional screening methods involve fundus cameras, which are costly and bulky, limiting their accessibility in remote or underserved areas.</p> Objective <p>This study aims to develop an automated DR detection model using smartphone-based retinal imaging, leveraging lightweight deep learning architectures to reduce blindness and complications associated with DR. The approach focuses on affordability, portability, and accessibility to enable non-specialists to capture and analyze retinal images outside traditional medical environments.</p> Methods <p>The proposed method utilizes a modified Mobilenet deep learning network by removing the full connection structure, which significantly reduces testing time. The model requires only 8 seconds for testing, outperforming traditional built-in classifiers in Mobilenet and the improved Mobilenet+SVM model (9.56 seconds testing time). Retinal images are captured using smartphone-compatible adapters, and the model is trained and tested on two publicly available datasets: EyePACS and APTOS-2019-Blindness-Detection. Performance is evaluated using metrics such as accuracy, sensitivity, specificity, and F1-score to ensure a comprehensive comparison with existing methods.</p> Results <p>The proposed model achieves an overall classification accuracy of 90%, surpassing the accuracy of traditional methods and other smartphone-based devices. The lightweight architecture ensures high performance with reduced computational requirements, making it more efficient for real-time applications.</p> Conclusion <p>The study highlights the potential of smartphone-based retinal imaging systems as a cost-effective and accessible alternative for DR detection. By leveraging the increasing prevalence of smartphones and advancements in deep learning, the proposed approach facilitates early DR screening in remote and underserved areas, addressing the limitations of traditional methods.</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

Automated detection of diabetic retinopathy using an improved deep learning model with smartphone images

  • 期刊论文

  • 发表日期：

    2025年1月23日

查看全文
  • Usharani Bhimavarapu

摘要

Background

Diabetic retinopathy (DR) is a diabetes-induced retinal disease characterized by damage to the blood vessels of the retina. If left undiagnosed, DR can lead to vision impairment or blindness. The asymptomatic nature of the disease, often referred to as a “silent disease,” makes early detection critical. Traditional screening methods involve fundus cameras, which are costly and bulky, limiting their accessibility in remote or underserved areas.

Objective

This study aims to develop an automated DR detection model using smartphone-based retinal imaging, leveraging lightweight deep learning architectures to reduce blindness and complications associated with DR. The approach focuses on affordability, portability, and accessibility to enable non-specialists to capture and analyze retinal images outside traditional medical environments.

Methods

The proposed method utilizes a modified Mobilenet deep learning network by removing the full connection structure, which significantly reduces testing time. The model requires only 8 seconds for testing, outperforming traditional built-in classifiers in Mobilenet and the improved Mobilenet+SVM model (9.56 seconds testing time). Retinal images are captured using smartphone-compatible adapters, and the model is trained and tested on two publicly available datasets: EyePACS and APTOS-2019-Blindness-Detection. Performance is evaluated using metrics such as accuracy, sensitivity, specificity, and F1-score to ensure a comprehensive comparison with existing methods.

Results

The proposed model achieves an overall classification accuracy of 90%, surpassing the accuracy of traditional methods and other smartphone-based devices. The lightweight architecture ensures high performance with reduced computational requirements, making it more efficient for real-time applications.

Conclusion

The study highlights the potential of smartphone-based retinal imaging systems as a cost-effective and accessible alternative for DR detection. By leveraging the increasing prevalence of smartphones and advancements in deep learning, the proposed approach facilitates early DR screening in remote and underserved areas, addressing the limitations of traditional methods.