Scholarlink

<p>Non-mydriatic retinal images, akin to neuroimaging, often suffer from artifacts and blurring that significantly inhibit the accurate diagnosis and early detection of neurodegenerative diseases. Current image enhancement methods, such as Robust Principal Component Analysis (RPCA), are commonly employed for early disease detection and medical diagnosis. However, these methods typically assume uniform singular value weights with existing nuclear norms, an assumption that may not hold because of the inherent variations in noisy image data. In addition, RPCA approaches primarily focus on global enhancement, failing to capture the fine details critical for precise analysis of retinal images. This limitation underscores the need for advanced and more adaptive method that can address these challenges and improve the accuracy of diagnostic tools in the biosciences, particularly for the early detection of diseases. To address these drawbacks, we propose a novel method that combines RPCA, the truncated weighted nuclear norm (TWNN), and the equalization of adaptive histograms (AHE). Unlike existing methods, the proposed approach enhances degraded retinal images by further applying Gaussian filtering to correct the edges and AHE used to detail the retinal images, preserve key structures, and improve contrast. The method is formulated as an optimization technique, incorporating Histogram Oriented Gradients (HOG) features, with the enhanced images used for diabetes prediction via machine learning, and the parameters are updated iteratively using ADMM. We conducted ablation studies to select the best model that predicts diabetes from the machine learning algorithms, including Support Vector Machine, Logistic Regression, and K-Nearest Neighbors, on retinal fundus images across all severity levels using the proposed method. Among these, SVM demonstrated the highest performance in predicting diabetes, from which we considered it for the classification purpose in this study. .One of the interesting aspect of this study is that, unlike the existing methods, we use the enhanced images produced by our approach for diabetes classification.The results of the study show that the proposed method substantially enhanced the quality of the images and improved prediction accuracy based on the public databases.</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

Robust PCA with Machine Learning for Retinal Image Enhancement and Diabetes Classification

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月23日

查看全文
  • Habte Tadesse Likassa,
  • Ding-Geng Chen

摘要

Non-mydriatic retinal images, akin to neuroimaging, often suffer from artifacts and blurring that significantly inhibit the accurate diagnosis and early detection of neurodegenerative diseases. Current image enhancement methods, such as Robust Principal Component Analysis (RPCA), are commonly employed for early disease detection and medical diagnosis. However, these methods typically assume uniform singular value weights with existing nuclear norms, an assumption that may not hold because of the inherent variations in noisy image data. In addition, RPCA approaches primarily focus on global enhancement, failing to capture the fine details critical for precise analysis of retinal images. This limitation underscores the need for advanced and more adaptive method that can address these challenges and improve the accuracy of diagnostic tools in the biosciences, particularly for the early detection of diseases. To address these drawbacks, we propose a novel method that combines RPCA, the truncated weighted nuclear norm (TWNN), and the equalization of adaptive histograms (AHE). Unlike existing methods, the proposed approach enhances degraded retinal images by further applying Gaussian filtering to correct the edges and AHE used to detail the retinal images, preserve key structures, and improve contrast. The method is formulated as an optimization technique, incorporating Histogram Oriented Gradients (HOG) features, with the enhanced images used for diabetes prediction via machine learning, and the parameters are updated iteratively using ADMM. We conducted ablation studies to select the best model that predicts diabetes from the machine learning algorithms, including Support Vector Machine, Logistic Regression, and K-Nearest Neighbors, on retinal fundus images across all severity levels using the proposed method. Among these, SVM demonstrated the highest performance in predicting diabetes, from which we considered it for the classification purpose in this study. .One of the interesting aspect of this study is that, unlike the existing methods, we use the enhanced images produced by our approach for diabetes classification.The results of the study show that the proposed method substantially enhanced the quality of the images and improved prediction accuracy based on the public databases.