Scholarlink

<p>Colorectal cancer (CRC) is the second leading cause of cancer-related deaths, highlighting the need for early detection. CRC often starts as polyps. Most polyps are benign, although some can evolve and eventually become cancerous. While colonoscopy is the gold standard for polyp detection, Computed Tomography CT colonography offers a non-invasive alternative, yet CRC classification with 3D CT data is underexplored. In this context, this study introduces 3D CTCAD, a Computer-Aided Detection (CAD) system using 3D Convolutional Neural Network (3D-CNN) and 3DAlexNet models to improve polyp detection in CT colonography images. A significant challenge posed is the imbalanced nature of the dataset, which can cause the model to prioritize the majority class and thus hinder the system's performance. To tackle this issue, we addressed dataset imbalance by applying the Synthetic Minority Oversampling Technique (SMOTE) to the ACRIN CT COLONOGRAPHY dataset. We explored both a standard data split of 70:15:15 (train:validation:test) and an optimized split, based on the number of samples in the dataset, of 91:04:05. The 3D-CNN model, with SMOTE, reached a recall of 92.90% and an accuracy of 89%. While 3DAlexNet has a recall of 92.30% and an accuracy of 92%. Previous studies used 2D input and pre-trained CNN architectures, achieving a recall ranging from 80%- 99% and an accuracy rate between 90.1% and 98.75%. Our 3D approach yielded competitive results, highlighting that even with training from scratch, our 3D models can rival pre-trained 2D models. These findings suggest that 3D CTCAD could enhance non-invasive CRC screening, supporting early detection and complementing current diagnostic tools.</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

3D-CTCAD: a novel robust system on colorectal cancer prevention based on optimal split approach

  • 期刊论文

  • 发表日期：

    2025年4月4日

查看全文
  • Khadija Hicham,
  • Sara Laghmati,
  • Soufiane Hamida,
  • Amal Tmiri,
  • Bouchaib Cherradi

摘要

Colorectal cancer (CRC) is the second leading cause of cancer-related deaths, highlighting the need for early detection. CRC often starts as polyps. Most polyps are benign, although some can evolve and eventually become cancerous. While colonoscopy is the gold standard for polyp detection, Computed Tomography CT colonography offers a non-invasive alternative, yet CRC classification with 3D CT data is underexplored. In this context, this study introduces 3D CTCAD, a Computer-Aided Detection (CAD) system using 3D Convolutional Neural Network (3D-CNN) and 3DAlexNet models to improve polyp detection in CT colonography images. A significant challenge posed is the imbalanced nature of the dataset, which can cause the model to prioritize the majority class and thus hinder the system's performance. To tackle this issue, we addressed dataset imbalance by applying the Synthetic Minority Oversampling Technique (SMOTE) to the ACRIN CT COLONOGRAPHY dataset. We explored both a standard data split of 70:15:15 (train:validation:test) and an optimized split, based on the number of samples in the dataset, of 91:04:05. The 3D-CNN model, with SMOTE, reached a recall of 92.90% and an accuracy of 89%. While 3DAlexNet has a recall of 92.30% and an accuracy of 92%. Previous studies used 2D input and pre-trained CNN architectures, achieving a recall ranging from 80%- 99% and an accuracy rate between 90.1% and 98.75%. Our 3D approach yielded competitive results, highlighting that even with training from scratch, our 3D models can rival pre-trained 2D models. These findings suggest that 3D CTCAD could enhance non-invasive CRC screening, supporting early detection and complementing current diagnostic tools.