Scholarlink

<p>This narrative review explores advanced Artificial Intelligence (AI) tools, particularly Convolutional Neural Network (CNN), for automated fish disease diagnosis, including key technologies, clinical applications, ethical constraints, and future insights. Given that fish disease diagnosis is essential for the aquaculture industry and that the diagnostic tools are costly, it was imperative to employ Artificial Intelligence (AI) to automate fish disease management. Within this context, the CNN-based analysis has been integrated into fish disease diagnosis, suggesting its key role in improving disease management practices in different aquaculture systems. This integration enhances practitioners’ and researchers’ knowledge, understanding, and advances their ability to improve management practices within aquaculture systems. This review presents modern tools based on CNN models for aquaculture, including image acquisition, preprocessing, segmentation, feature extraction, classification, transfer learning, and deployment. Additionally, it highlights broader applications of computer vision in aquaculture, the performance of the outputs, and the challenges that limit real-world implementation. These include poor data quality, class imbalance, domain shift, overfitting, limited interpretability, uncertainty in model predictions, reduced robustness under field imaging conditions, and the need for continuous human supervision. However, many studies have reported encouraging experimental outcomes; systems based on CNNs have not yet been investigated across different farm settings, imaging conditions, and disease stages. Thus, CNNs should be considered earlier diagnostic and supportive decision tools rather than a replacement for veterinary diagnosis or laboratory confirmation. These AI models have been trained and validated; however, they may still not represent the farming environment variability. Therefore, we were keen to address these limitations, which are essential to translating experimental success into practical disease management.</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 fish disease diagnosis in aquaculture using convolutional neural networks: a narrative review of methods, applications, and challenges

  • 期刊论文

  • 发表日期：

    2026年7月8日

查看全文
  • Humza Shaikh,
  • Al-Zahraa Mamdouh,
  • Fatma Ahmed,
  • Eman Zahran,
  • Ahmed Mustafa

摘要

This narrative review explores advanced Artificial Intelligence (AI) tools, particularly Convolutional Neural Network (CNN), for automated fish disease diagnosis, including key technologies, clinical applications, ethical constraints, and future insights. Given that fish disease diagnosis is essential for the aquaculture industry and that the diagnostic tools are costly, it was imperative to employ Artificial Intelligence (AI) to automate fish disease management. Within this context, the CNN-based analysis has been integrated into fish disease diagnosis, suggesting its key role in improving disease management practices in different aquaculture systems. This integration enhances practitioners’ and researchers’ knowledge, understanding, and advances their ability to improve management practices within aquaculture systems. This review presents modern tools based on CNN models for aquaculture, including image acquisition, preprocessing, segmentation, feature extraction, classification, transfer learning, and deployment. Additionally, it highlights broader applications of computer vision in aquaculture, the performance of the outputs, and the challenges that limit real-world implementation. These include poor data quality, class imbalance, domain shift, overfitting, limited interpretability, uncertainty in model predictions, reduced robustness under field imaging conditions, and the need for continuous human supervision. However, many studies have reported encouraging experimental outcomes; systems based on CNNs have not yet been investigated across different farm settings, imaging conditions, and disease stages. Thus, CNNs should be considered earlier diagnostic and supportive decision tools rather than a replacement for veterinary diagnosis or laboratory confirmation. These AI models have been trained and validated; however, they may still not represent the farming environment variability. Therefore, we were keen to address these limitations, which are essential to translating experimental success into practical disease management.