Scholarlink

<p>Cancer seems to be a leading global health concern, responsible for millions of deaths each year due to late diagnosis and limited access to advanced diagnosis tools. Early and precise detection of cancer is critical for improving patient’s survival and treatment outcomes. To address these challenges, this study proposes a robust deep learning-based framework for multi-cancer detection, targeting brain, breast and lung cancers. A real-time MRI dataset was collected from Capital Hospital, Bhubaneswar, India, ensuring clinical authenticity and diversity. To mitigate the limitation of small dataset, a comprehensive preprocessing and data synthesis pipeline was implemented; incorporating image enhancement and MedGAN based augmentation to expand variability and improve model generalizability. A DL model, VGG-19 was employed for feature extraction, integrating transformer modules to capture rich contextual and spatial representations from MRI scans. The extracted features were further optimized using the Adaptive genetic Algorithm (AGA) to enhance detection accuracy, reduce prediction error and ensure convergence stability. The model robustness was evaluated through 9-fold cross validation and additional 5% external dataset testing confirming generalization across diverse patient sample. The proposed framework demonstrated excellent performance achieving an accuracy of 98.7 ± 0.34%, precision of 0.973, recall of 0.967 and F1 score of 0.981. For interpretability Grad-CAM visualization was applied, providing clinically meaningful insights into the model’s decision-making process. This framework highlights the potential of combining advanced feature extraction, transformer integration, metaheuristic optimization and rigorous validation to build a scalable and reliable AI-assisted diagnostic tool, capable of accurate and interpretable multi cancer detection for real world clinical applications.</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

Enhanced Multi-Cancer Diagnosis Using a Fine-Tuned Transformer-VGG19 Approach

  • 期刊论文

  • 发表日期：

    2026年1月7日

查看全文
  • Kumar Janardan Patra,
  • Jibitesh Mishra,
  • Debabrata Singh

摘要

Cancer seems to be a leading global health concern, responsible for millions of deaths each year due to late diagnosis and limited access to advanced diagnosis tools. Early and precise detection of cancer is critical for improving patient’s survival and treatment outcomes. To address these challenges, this study proposes a robust deep learning-based framework for multi-cancer detection, targeting brain, breast and lung cancers. A real-time MRI dataset was collected from Capital Hospital, Bhubaneswar, India, ensuring clinical authenticity and diversity. To mitigate the limitation of small dataset, a comprehensive preprocessing and data synthesis pipeline was implemented; incorporating image enhancement and MedGAN based augmentation to expand variability and improve model generalizability. A DL model, VGG-19 was employed for feature extraction, integrating transformer modules to capture rich contextual and spatial representations from MRI scans. The extracted features were further optimized using the Adaptive genetic Algorithm (AGA) to enhance detection accuracy, reduce prediction error and ensure convergence stability. The model robustness was evaluated through 9-fold cross validation and additional 5% external dataset testing confirming generalization across diverse patient sample. The proposed framework demonstrated excellent performance achieving an accuracy of 98.7 ± 0.34%, precision of 0.973, recall of 0.967 and F1 score of 0.981. For interpretability Grad-CAM visualization was applied, providing clinically meaningful insights into the model’s decision-making process. This framework highlights the potential of combining advanced feature extraction, transformer integration, metaheuristic optimization and rigorous validation to build a scalable and reliable AI-assisted diagnostic tool, capable of accurate and interpretable multi cancer detection for real world clinical applications.