Scholarlink

The use of Artificial Intelligence (AI) in the medical field has grown significantly in recent years, also with computer vision techniques, such as Convolutional Neural Networks (CNNs), gaining prominence in medical image processing applications. While CNNs have proven helpful in various medical tasks, there is a research gap in the automatic selection of Corneal Confocal Microscopy (CCM) images for diagnostic purposes. CCM could be used to screen for early stages of Diabetic Peripheral Neuropathy (DPN), the most common complication of diabetes. However, manually selecting CCM images for diagnosis is time-consuming and requires expertise. This paper aims to evaluate the effectiveness of a Genetic Algorithm (GA) in finding an optimized CNN architecture for accurate quality control of CCM images. The GA-based approach is compared to the widely used VGG16 architecture. The results demonstrate that the GA-evolved CNN architecture achieved a validation accuracy of 67.57 to 72.97%, outperforming VGG16. This research highlights the potential of GAs in optimizing CNN architectures for medical image classification tasks and indicates a first step towards automatically selecting CCM images for analysis.

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

Optimizing CNN Architecture for Quality Control of Corneal Confocal Microscopy Images Using a Genetic Algorithm

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Berenice N. Bönhof,
  • Thomas Hanne,
  • Rolf Dornberger,
  • Phillip Gachnang,
  • Gidon J. Bönhof

摘要

The use of Artificial Intelligence (AI) in the medical field has grown significantly in recent years, also with computer vision techniques, such as Convolutional Neural Networks (CNNs), gaining prominence in medical image processing applications. While CNNs have proven helpful in various medical tasks, there is a research gap in the automatic selection of Corneal Confocal Microscopy (CCM) images for diagnostic purposes. CCM could be used to screen for early stages of Diabetic Peripheral Neuropathy (DPN), the most common complication of diabetes. However, manually selecting CCM images for diagnosis is time-consuming and requires expertise. This paper aims to evaluate the effectiveness of a Genetic Algorithm (GA) in finding an optimized CNN architecture for accurate quality control of CCM images. The GA-based approach is compared to the widely used VGG16 architecture. The results demonstrate that the GA-evolved CNN architecture achieved a validation accuracy of 67.57 to 72.97%, outperforming VGG16. This research highlights the potential of GAs in optimizing CNN architectures for medical image classification tasks and indicates a first step towards automatically selecting CCM images for analysis.