Scholarlink

This study introduces a two-stage deep learning system with a view to increasing the accuracy of brain tumors that detect brain tumors using CT scan images. In the first phase, the image resolution is improved through a generative adversarial network for super resolution (SR-GAN), which allows the extraction of fine structural properties. The classification is then performed using a hybrid model that adds ResNet101 and EfficientNet-B5 models, which benefit from the efficiency of the scaling for the depth of the residuals and better diagnostic precision. The second phase uses YOLOv11 as the algorithm to detect a real-time object, where the tumor areas are accurately located in images. The CT Analysis is designed via normalization, preprocessing, and development to improve the normalization and stability of the dataset model. The system is evaluated on the use of preferred metrics, together with accuracy, precision, and F1-score average (mAP). The hybrid classification model receives an impressive 98% accuracy, while YOLOv11 indicates a dependable localization with minimum false positivity. This integrated method not only improves detection rate but also supports doctors in informed decision-making. The overall system provides robust, automatic equipment that can help in early diagnosis and effective treatment schemes within the healthcare system.

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

A Multi-Model Approach for Brain Tumor Detection in CT Scans Using YOLOv11 and Pre-trained Networks

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Sunil Babu Melingi,
  • Kavya Sree Tota,
  • Jeevan Preethi Burramukku,
  • Mohan Sai Thota,
  • Venkateswara Rao Uppala

摘要

This study introduces a two-stage deep learning system with a view to increasing the accuracy of brain tumors that detect brain tumors using CT scan images. In the first phase, the image resolution is improved through a generative adversarial network for super resolution (SR-GAN), which allows the extraction of fine structural properties. The classification is then performed using a hybrid model that adds ResNet101 and EfficientNet-B5 models, which benefit from the efficiency of the scaling for the depth of the residuals and better diagnostic precision. The second phase uses YOLOv11 as the algorithm to detect a real-time object, where the tumor areas are accurately located in images. The CT Analysis is designed via normalization, preprocessing, and development to improve the normalization and stability of the dataset model. The system is evaluated on the use of preferred metrics, together with accuracy, precision, and F1-score average (mAP). The hybrid classification model receives an impressive 98% accuracy, while YOLOv11 indicates a dependable localization with minimum false positivity. This integrated method not only improves detection rate but also supports doctors in informed decision-making. The overall system provides robust, automatic equipment that can help in early diagnosis and effective treatment schemes within the healthcare system.