Scholarlink

Background <p>Diabetic retinopathy (DR) and diabetic macular edema are significant causes of visual loss among working adults. Early detection and classification are essential for effective management. Traditional diagnostic methods involve fundus biomicroscopy or grading of fundus photography, but artificial intelligence (AI)-driven automated image recognition systems have emerged as valuable diagnostic tools.</p> Objective <p>This study aims to evaluate the performance and efficacy of the Viderai software, an AI-driven system designed for the detection and classification of DR from high-quality color fundus photographs. The results from Viderai were compared to assessments by expert consultant retinologists.</p> Methods <p>A prospective, double-blinded clinical trial was conducted at University Hospital Ostrava between June and October 2023. A total of 1214 fundus photographs were captured from 615 patients. After exclusions, 1170 images were used for validity assessment, and 918 images were analyzed for DR detection and classification. Statistical analyses, including sensitivity, specificity, precision, and F1 score calculations, were performed to compare Viderai’s performance with that of retinal specialists.</p> Results <p>The agreement between Viderai and physicians in image validity assessment was 91%, with 86% of images deemed suitable for DR evaluation. Sensitivity, specificity, accuracy, and F1 scores for validity assessment were 90.7%, 94.9%, 99.1%, and 94.7% (any match) and 77.8%, 47.9%, 80.4%, and 79.1% (exact match), respectively. In DR classification, the agreement rate was 95% for presence/absence and 93% for classification accuracy. Sensitivity, specificity, accuracy, and F1 scores for DR classification were 97.2%, 94.3%, 86.3%, and 91.4% (any match) and 89.6%, 93.6%, 83.4%, and 86.4% (exact match), respectively.</p> Conclusion <p>The study demonstrates that Viderai is a reliable tool for DR detection and classification, with performance metrics comparable to existing AI-based systems. The findings support the potential for Viderai’s integration into clinical practice, aiding in early DR diagnosis and management. Further studies are warranted to explore its broader applicability.</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 Diabetic Retinopathy Detection and Classification from Fundus Photography in Czech Republic Population: Prospective Comparative Study

  • 期刊论文

  • 发表日期：

    2025年9月8日

查看全文
  • Jan Němčanský,
  • Jan Studnička,
  • Sabina Němčanská,
  • Michal Koubek,
  • Radka Nágelová,
  • Veronika Fialová,
  • Nora Majtanová,
  • Pavel Rozsíval

摘要

Background

Diabetic retinopathy (DR) and diabetic macular edema are significant causes of visual loss among working adults. Early detection and classification are essential for effective management. Traditional diagnostic methods involve fundus biomicroscopy or grading of fundus photography, but artificial intelligence (AI)-driven automated image recognition systems have emerged as valuable diagnostic tools.

Objective

This study aims to evaluate the performance and efficacy of the Viderai software, an AI-driven system designed for the detection and classification of DR from high-quality color fundus photographs. The results from Viderai were compared to assessments by expert consultant retinologists.

Methods

A prospective, double-blinded clinical trial was conducted at University Hospital Ostrava between June and October 2023. A total of 1214 fundus photographs were captured from 615 patients. After exclusions, 1170 images were used for validity assessment, and 918 images were analyzed for DR detection and classification. Statistical analyses, including sensitivity, specificity, precision, and F1 score calculations, were performed to compare Viderai’s performance with that of retinal specialists.

Results

The agreement between Viderai and physicians in image validity assessment was 91%, with 86% of images deemed suitable for DR evaluation. Sensitivity, specificity, accuracy, and F1 scores for validity assessment were 90.7%, 94.9%, 99.1%, and 94.7% (any match) and 77.8%, 47.9%, 80.4%, and 79.1% (exact match), respectively. In DR classification, the agreement rate was 95% for presence/absence and 93% for classification accuracy. Sensitivity, specificity, accuracy, and F1 scores for DR classification were 97.2%, 94.3%, 86.3%, and 91.4% (any match) and 89.6%, 93.6%, 83.4%, and 86.4% (exact match), respectively.

Conclusion

The study demonstrates that Viderai is a reliable tool for DR detection and classification, with performance metrics comparable to existing AI-based systems. The findings support the potential for Viderai’s integration into clinical practice, aiding in early DR diagnosis and management. Further studies are warranted to explore its broader applicability.