Scholarlink

<p>Diabetic Retinopathy (DR) is a leading cause of vision-threatening conditions worldwide. In clinical settings, the automated identification of DR using Multicolor Imaging (MCI) is critical for assisting ophthalmologists in the timely diagnosis and management of the condition. Deep Learning (DL) methods have been developed to automatically grade DR, enabling ophthalmologists to design personalized treatment plans for patients. However, a significant research gap remains in the application of DL for MCI analysis, particularly in leveraging its multimodal feature representations. This study proposes a novel approach, the Multimodal Network Incorporating Information Bottleneck (MNIIB), specifically designed for DR classification using MCI. Unlike previous approaches that primarily apply Information Bottleneck (IB) theory to model input-label relationships, our MNIIB framework explicitly employs IB principles to analyze and optimize the interactions between the different imaging modalities within MCI. It integrates features extracted from multiple modalities and incorporates an IB mechanism to identify and refine the shared information across modalities, thereby compressing redundant data and enhancing the extraction of diagnostically relevant features. The effectiveness of the proposed network was validated through extensive testing on a retinal image dataset, achieving an accuracy of 95.9%. These results highlight the potential of the MNIIB as a reliable diagnostic tool for the early and accurate detection of DR.</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

Detection of diabetic retinopathy using multicolor image by multimodal network incorporating information bottleneck (MNIIB)

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月14日

查看全文
  • Jingqi Song,
  • Cuihuan Tian,
  • Yining Qi,
  • Xinke Gao,
  • Min Cui,
  • Jixin Yang,
  • Kexue Wu,
  • Xiaoming Xi

摘要

Diabetic Retinopathy (DR) is a leading cause of vision-threatening conditions worldwide. In clinical settings, the automated identification of DR using Multicolor Imaging (MCI) is critical for assisting ophthalmologists in the timely diagnosis and management of the condition. Deep Learning (DL) methods have been developed to automatically grade DR, enabling ophthalmologists to design personalized treatment plans for patients. However, a significant research gap remains in the application of DL for MCI analysis, particularly in leveraging its multimodal feature representations. This study proposes a novel approach, the Multimodal Network Incorporating Information Bottleneck (MNIIB), specifically designed for DR classification using MCI. Unlike previous approaches that primarily apply Information Bottleneck (IB) theory to model input-label relationships, our MNIIB framework explicitly employs IB principles to analyze and optimize the interactions between the different imaging modalities within MCI. It integrates features extracted from multiple modalities and incorporates an IB mechanism to identify and refine the shared information across modalities, thereby compressing redundant data and enhancing the extraction of diagnostically relevant features. The effectiveness of the proposed network was validated through extensive testing on a retinal image dataset, achieving an accuracy of 95.9%. These results highlight the potential of the MNIIB as a reliable diagnostic tool for the early and accurate detection of DR.