Scholarlink

<p>Medical image analysis plays a vital role in diagnosing diseases and assisting healthcare professionals in providing accurate and timely treatment. Convolutional neural networks (CNNs) have proven to be highly effective in extracting relevant features from medical images. However, due to the limited availability of annotated medical datasets, training CNNs from scratch becomes challenging. Transfer Learning offers a promising solution to address this issue by leveraging the pre-trained models on large-scale datasets. Numerous studies have aimed to conduct a comparative analysis of medical datasets using Transfer Learning. These studies involve training and evaluating multiple CNN architectures, such as VGGs, ResNet, and Inception, with varying degrees of transfer learning. However, the existing methodology often focuses on fine-tuning the entire network or the classifier layer, neglecting the individual blocks within the CNN architectures. This approach requires updating the entire network, which can be computationally expensive and may lead to overfitting when working with limited target datasets. We propose a novel paradigm called ’Single Block Over Network’ (SBON) for fine-tuning neural networks, which focuses on selectively updating a single block of pre-trained CNN architectures to enhance model adaptation to specific features. We also intend to study the effect of Optimizers and SOTA CNNs &amp; Transformers on Medical Datasets. The effectiveness of our SBON approach and architectures will be verified on two benchmark datasets: Chest X-Ray Pneumonia and RSNA Breast Cancer Detection Dataset, widely recognized in the medical imaging community for their importance in diagnostic research. Our experimental results indicate that our proposed strategy offers a promising approach for fine-tuning a single component of a pre-trained network in the context of medical analysis and datasets.</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

Single block over network (SBON): a paradigm for efficient fine-tuning of neural networks in medical imaging

  • 期刊论文

  • 发表日期：

    2026年2月3日

查看全文
  • Utsav Khatu,
  • Pranit Rohokale,
  • Sahil Kedare,
  • Aryan Devikar,
  • Sandip Shingade,
  • Sushila Shingade

摘要

Medical image analysis plays a vital role in diagnosing diseases and assisting healthcare professionals in providing accurate and timely treatment. Convolutional neural networks (CNNs) have proven to be highly effective in extracting relevant features from medical images. However, due to the limited availability of annotated medical datasets, training CNNs from scratch becomes challenging. Transfer Learning offers a promising solution to address this issue by leveraging the pre-trained models on large-scale datasets. Numerous studies have aimed to conduct a comparative analysis of medical datasets using Transfer Learning. These studies involve training and evaluating multiple CNN architectures, such as VGGs, ResNet, and Inception, with varying degrees of transfer learning. However, the existing methodology often focuses on fine-tuning the entire network or the classifier layer, neglecting the individual blocks within the CNN architectures. This approach requires updating the entire network, which can be computationally expensive and may lead to overfitting when working with limited target datasets. We propose a novel paradigm called ’Single Block Over Network’ (SBON) for fine-tuning neural networks, which focuses on selectively updating a single block of pre-trained CNN architectures to enhance model adaptation to specific features. We also intend to study the effect of Optimizers and SOTA CNNs & Transformers on Medical Datasets. The effectiveness of our SBON approach and architectures will be verified on two benchmark datasets: Chest X-Ray Pneumonia and RSNA Breast Cancer Detection Dataset, widely recognized in the medical imaging community for their importance in diagnostic research. Our experimental results indicate that our proposed strategy offers a promising approach for fine-tuning a single component of a pre-trained network in the context of medical analysis and datasets.