Scholarlink

<p>A significant challenge in medical healthcare is the limited availability of labelled imaging data, coupled with class imbalance issues, which adversely affect the performance of learning algorithms reliant on large, high-quality datasets. In many cases, datasets contain unequal disease and non-disease cases, leading to biased models that struggle with generalisation. Therefore, this research explores the Deep Convolutional Generative Adversarial Networks (DCGANs) to generate synthetic chest X-ray images, dealing with the issue of class imbalance. The system is further trained using the ResNet50 model for the augmented chest X-ray image dataset labelled either pneumonia or normal. Evaluation using DCGANs with ResNet50 demonstrated a success rate of 93.4% in producing photorealistic images, validating their potential to augment existing datasets. The synthetic images facilitate improved model training, enhancing diagnostic accuracy in medical facilities by providing enriched medical imaging datasets. This study highlights how advanced generative methods can address healthcare data constraints, enabling precise and efficient diagnostic systems.</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

Mitigating risk in medical AI: balancing X-ray datasets for reliable detection

  • 期刊论文

  • 发表日期：

    2025年5月21日

查看全文
  • Mohd Aquib Ansari,
  • Shahnawaz Ahmad,
  • Arvind Mewada

摘要

A significant challenge in medical healthcare is the limited availability of labelled imaging data, coupled with class imbalance issues, which adversely affect the performance of learning algorithms reliant on large, high-quality datasets. In many cases, datasets contain unequal disease and non-disease cases, leading to biased models that struggle with generalisation. Therefore, this research explores the Deep Convolutional Generative Adversarial Networks (DCGANs) to generate synthetic chest X-ray images, dealing with the issue of class imbalance. The system is further trained using the ResNet50 model for the augmented chest X-ray image dataset labelled either pneumonia or normal. Evaluation using DCGANs with ResNet50 demonstrated a success rate of 93.4% in producing photorealistic images, validating their potential to augment existing datasets. The synthetic images facilitate improved model training, enhancing diagnostic accuracy in medical facilities by providing enriched medical imaging datasets. This study highlights how advanced generative methods can address healthcare data constraints, enabling precise and efficient diagnostic systems.