Scholarlink

The global incidence of prostate cancer has steadily increased over time, mainly due to the growing proportion of the older population. Early detection of prostate cancer, when it is still confined to the prostate gland, significantly enhances the chances of successful treatment and improves survival rates. This study proposes an automated prostate cancer diagnosis and segmentation system using advanced medical image processing techniques. The most relevant features from preprocessed prostate MRI images are selected using a novel Wild Horse Optimized (WHO) feature selection method to optimise the training and testing performance of the cancer detection system. A suite of deep learning models, including Convolutional Neural Network (CNN), Residual Network (ResNet), and Generative Adversarial Network (GAN), is then utilized to classify MRI images as cancerous or non-cancerous accurately. The final classification result is determined through a voting mechanism that selects the best-performing prediction model. Furthermore, a new technique, Dual Swin Transformer UNet Segmentation (DSTra-UNet), is introduced to precisely segment cancer-affected regions within the prostate images. The proposed system is evaluated using various performance metrics, including accuracy, sensitivity, precision, recall, and error rate, demonstrating its effectiveness and reliability in prostate cancer diagnosis and segmentation.

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

Ensemble-Based Automated Prostate Cancer Diagnosis Using Dual Swin Transformer UNet Segmentation (DSTra-UNet)

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • E. Sathesh Abraham Leo,
  • K. Nattar Kannan,
  • Gunasekar Thangarasu,
  • Saravanan Muthaiyah,
  • Thein Oak Kyaw Zaw

摘要

The global incidence of prostate cancer has steadily increased over time, mainly due to the growing proportion of the older population. Early detection of prostate cancer, when it is still confined to the prostate gland, significantly enhances the chances of successful treatment and improves survival rates. This study proposes an automated prostate cancer diagnosis and segmentation system using advanced medical image processing techniques. The most relevant features from preprocessed prostate MRI images are selected using a novel Wild Horse Optimized (WHO) feature selection method to optimise the training and testing performance of the cancer detection system. A suite of deep learning models, including Convolutional Neural Network (CNN), Residual Network (ResNet), and Generative Adversarial Network (GAN), is then utilized to classify MRI images as cancerous or non-cancerous accurately. The final classification result is determined through a voting mechanism that selects the best-performing prediction model. Furthermore, a new technique, Dual Swin Transformer UNet Segmentation (DSTra-UNet), is introduced to precisely segment cancer-affected regions within the prostate images. The proposed system is evaluated using various performance metrics, including accuracy, sensitivity, precision, recall, and error rate, demonstrating its effectiveness and reliability in prostate cancer diagnosis and segmentation.