Scholarlink

Thyroid cancer is one of the most common cancers in the world today, with women being more likely to suffer from this disease than men. One of the fundamentals for curing and predicting thyroid cancer is early detection and diagnosis. In this study, we propose a method for classifying thyroid cancer abnormalities using Convolutional Neural Networks (CNNs) combined with Support Vector Machines (SVMs) and data augmentation to help physicians diagnose thyroid-related diseases. Unlike previous studies that applied data augmentation without targeted optimization, we systematically examine the effects of rotation and cropping angles on classification accuracy, identifying optimal settings to improve performance. After experimental evaluation on the DS1 thyroid ultrasound image dataset, our method achieved an accuracy of 96.97%, a sensitivity of 98.6%, and an F-measure of 97.57%, which are higher than some published studies, which shows great promise in practical applications for classifying and detecting abnormalities in thyroid ultrasound image data.

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

Classification of Thyroid Nodule in Ultrasound Images Using CNN-SVM Hybrid with Data Augmentation

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Duong Trong Luong,
  • Dang Le Dat,
  • Nguyen Phuong Huy,
  • Duong Minh Khanh,
  • Tran Ngoc Tuan,
  • Do Duy Thai

摘要

Thyroid cancer is one of the most common cancers in the world today, with women being more likely to suffer from this disease than men. One of the fundamentals for curing and predicting thyroid cancer is early detection and diagnosis. In this study, we propose a method for classifying thyroid cancer abnormalities using Convolutional Neural Networks (CNNs) combined with Support Vector Machines (SVMs) and data augmentation to help physicians diagnose thyroid-related diseases. Unlike previous studies that applied data augmentation without targeted optimization, we systematically examine the effects of rotation and cropping angles on classification accuracy, identifying optimal settings to improve performance. After experimental evaluation on the DS1 thyroid ultrasound image dataset, our method achieved an accuracy of 96.97%, a sensitivity of 98.6%, and an F-measure of 97.57%, which are higher than some published studies, which shows great promise in practical applications for classifying and detecting abnormalities in thyroid ultrasound image data.