Scholarlink

Using a bilinear VGG16 architecture with transfer learning, we suggest a deep learning method for identifying benign or cancerous thyroid ultrasound pictures. To lessen class imbalance, preprocessing and data augmentation are used to the DDTI dataset. The suggested model shows promise as a diagnostic tool to lessen human error, achieving about 75% accuracy and excellent sensitivity for malignant patients. AUC-ROC, accuracy, recall, and precision are used to assess performance. The wider effects of deep learning in medical imaging are also examined in this paper, especially in relation to telemedicine and AI-powered clinical processes. Scalability and accessibility are guaranteed by the model training pipeline’s use of TensorFlow/Keras on Google Colab. Grad-CAM and other explainability technologies are recognised as crucial future approaches to boost physician trust. These improvements establish the model as a possible clinical support tool in addition to an academic contribution. Practical factors including integration with medical information systems, ethical deployment, and computing efficiency are also highlighted in the paper.

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

Thyroid Cancer Detection from Ultrasound Images Using Deep Learning (Bilinear VGG16)

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Anusha Ampavathi,
  • D. Manohar,
  • Irfan Mohammed Ahmed,
  • Sufiya Shareef,
  • M. Siddharth Reddy,
  • Suresh Babu Changalasetty

摘要

Using a bilinear VGG16 architecture with transfer learning, we suggest a deep learning method for identifying benign or cancerous thyroid ultrasound pictures. To lessen class imbalance, preprocessing and data augmentation are used to the DDTI dataset. The suggested model shows promise as a diagnostic tool to lessen human error, achieving about 75% accuracy and excellent sensitivity for malignant patients. AUC-ROC, accuracy, recall, and precision are used to assess performance. The wider effects of deep learning in medical imaging are also examined in this paper, especially in relation to telemedicine and AI-powered clinical processes. Scalability and accessibility are guaranteed by the model training pipeline’s use of TensorFlow/Keras on Google Colab. Grad-CAM and other explainability technologies are recognised as crucial future approaches to boost physician trust. These improvements establish the model as a possible clinical support tool in addition to an academic contribution. Practical factors including integration with medical information systems, ethical deployment, and computing efficiency are also highlighted in the paper.