Scholarlink

<p>Artificial Intelligence (AI) is one of the most advanced technologies today, accelerating productivity, efficiency, and effectiveness in various sectors, including healthcare. However, defining such productivity requires data availability within the intervention field. In contrast to its potential, biomedical data for healthcare sector innovation is one of the most difficult data to access due to privacy concerns. One of the ways to overcome such difficulty is the use of AI to ‘Augment Data’ that has the potential to revolutionize healthcare sector productivity. This research paper presents a comprehensive survey on Data Augmentation (DA) for biomedical data, with specific interest in Generative Adversarial Network (GAN)-driven learning methods. It analyzes DA methods concerning the characteristics of these methods, performance metrics used to evaluate the quality of the generated data, and the limitations of each method. Additionally, this work includes a comprehensive experimental analysis to evaluate the performance of GANs and Conditional GANs (CGANs) on different data sources and sizes from two perspectives, which include characteristics and performance metrics. As a result, the analysis of existing GANs and modified versions of GANs indicates that there is significant potential in using generative models in biomedical applications. Key findings from this survey identify two major challenges in achieving reliable augmented data—a) unstable training for GAN models and b) the need for more reliable evaluation metrics. Addressing these challenges will be crucial for developing a new generation of GAN models that can ensure reliable DA techniques, with minimal training data and learning iterations.</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

Leveraging AI for biomedical data augmentation: a comparative review of model characteristics, performance analysis, and future research directions

  • 期刊论文

  • 发表日期：

    2025年12月1日

查看全文
  • Salha Albehairi,
  • Samiya Khan,
  • Reem Alotaibi,
  • Nofe Alganmi,
  • Mohammad Patwary

摘要

Artificial Intelligence (AI) is one of the most advanced technologies today, accelerating productivity, efficiency, and effectiveness in various sectors, including healthcare. However, defining such productivity requires data availability within the intervention field. In contrast to its potential, biomedical data for healthcare sector innovation is one of the most difficult data to access due to privacy concerns. One of the ways to overcome such difficulty is the use of AI to ‘Augment Data’ that has the potential to revolutionize healthcare sector productivity. This research paper presents a comprehensive survey on Data Augmentation (DA) for biomedical data, with specific interest in Generative Adversarial Network (GAN)-driven learning methods. It analyzes DA methods concerning the characteristics of these methods, performance metrics used to evaluate the quality of the generated data, and the limitations of each method. Additionally, this work includes a comprehensive experimental analysis to evaluate the performance of GANs and Conditional GANs (CGANs) on different data sources and sizes from two perspectives, which include characteristics and performance metrics. As a result, the analysis of existing GANs and modified versions of GANs indicates that there is significant potential in using generative models in biomedical applications. Key findings from this survey identify two major challenges in achieving reliable augmented data—a) unstable training for GAN models and b) the need for more reliable evaluation metrics. Addressing these challenges will be crucial for developing a new generation of GAN models that can ensure reliable DA techniques, with minimal training data and learning iterations.