Scholarlink

This study provides a thorough exploration of Vision Transformer (ViT) models in the realm of medical image analysis, specifically focusing on datasets with smaller size and greater complexity than traditional ViT evaluations. Demonstrating the transformative potential of ViT, our research highlights its capacity to enhance diagnostic accuracy and predictive capabilities in healthcare. These analyses were performed on Google Colaboratory, leveraging its cloud-based infrastructure for computational efficiency and accessibility. In the Brain Tumor model, ViT excelled with an accuracy of 85.71%, surpassing the 82.05% achieved by its CNN counterpart. ViT exhibited superior precision, recall, and F1-score, registering 0.88, 0.83, and 0.85, respectively, compared to the CNN model's 0.84, 0.79, and 0.81. The Lung Cancer Model similarly achieved an accuracy of 89.02% as compared to 85.05% and 83.01% compared to CNN models which showcases ViT's effectiveness in diverse medical imaging applications. Our work signifies a significant step towards leveraging ViT models in real-world medical scenarios, where complex data and nuanced patterns demand advanced analytical tools for improved patient outcomes.

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

From Pixels to Predictions: Medical Image Analysis and Prediction of Brain Tumor and Chest Cancer with Vision Transformers

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Bhagyashree Hambarde,
  • Mohammed Zubair Hussain,
  • Mohammed Saad Hussain,
  • Shrey Gupta,
  • Revanta Tomar,
  • Sanskriti Gupta

摘要

This study provides a thorough exploration of Vision Transformer (ViT) models in the realm of medical image analysis, specifically focusing on datasets with smaller size and greater complexity than traditional ViT evaluations. Demonstrating the transformative potential of ViT, our research highlights its capacity to enhance diagnostic accuracy and predictive capabilities in healthcare. These analyses were performed on Google Colaboratory, leveraging its cloud-based infrastructure for computational efficiency and accessibility. In the Brain Tumor model, ViT excelled with an accuracy of 85.71%, surpassing the 82.05% achieved by its CNN counterpart. ViT exhibited superior precision, recall, and F1-score, registering 0.88, 0.83, and 0.85, respectively, compared to the CNN model's 0.84, 0.79, and 0.81. The Lung Cancer Model similarly achieved an accuracy of 89.02% as compared to 85.05% and 83.01% compared to CNN models which showcases ViT's effectiveness in diverse medical imaging applications. Our work signifies a significant step towards leveraging ViT models in real-world medical scenarios, where complex data and nuanced patterns demand advanced analytical tools for improved patient outcomes.