Scholarlink

<p>Cervical cancer remains a critical public health challenge, especially in developing countries where mortality rates are alarmingly high. Accurate classification of cervical cancer cells is crucial due to the inherent complexity and variability in cellular images. Single models often struggle to fully capture the intricate patterns present in complex cell images, necessitating the use of ensemble methods to combine the collective strengths of various models. While previous models have utilized ensemble methods, there is still a significant demand for improved classification performance. This study presents an optimized ensemble method known as Cervi-Net, specifically designed to enhance the classification of cervical cancer cells. Our Cervi-Net model adeptly integrates three top performing well known pre-trained convolutional neural network (CNN) models including DenseNet169, MobileNetV2, and DenseNet201 to make them as base models. Importantly, we utilize grid search to intelligently assign weights to each base model to maximize their contributions for enhancing their collective accuracy. Evaluating the Cervi-Net model on the Mendeley LBC cervical Benchmark dataset (4-classes) reveals a remarkable overall diagnostic accuracy of 97.94%. Furthermore, we employ visualization techniques to highlight infected areas in test images, while ROC plots further assess the performance of proposed model. Additionally, statistical validation through the McNemar test supports the robustness of the model. These results demonstrate that Cervi-Net is a reliable and effective tool for assisting medical professionals in the early and accurate diagnosis of cervical cancer.</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

Automated cervical cancer cell diagnosis via grid search-optimized multi-CNN ensemble networks

  • 期刊论文

  • 发表日期：

    2025年7月18日

查看全文
  • Omair Bilal,
  • Arash Hekmat,
  • Saif Ur Rehman Khan

摘要

Cervical cancer remains a critical public health challenge, especially in developing countries where mortality rates are alarmingly high. Accurate classification of cervical cancer cells is crucial due to the inherent complexity and variability in cellular images. Single models often struggle to fully capture the intricate patterns present in complex cell images, necessitating the use of ensemble methods to combine the collective strengths of various models. While previous models have utilized ensemble methods, there is still a significant demand for improved classification performance. This study presents an optimized ensemble method known as Cervi-Net, specifically designed to enhance the classification of cervical cancer cells. Our Cervi-Net model adeptly integrates three top performing well known pre-trained convolutional neural network (CNN) models including DenseNet169, MobileNetV2, and DenseNet201 to make them as base models. Importantly, we utilize grid search to intelligently assign weights to each base model to maximize their contributions for enhancing their collective accuracy. Evaluating the Cervi-Net model on the Mendeley LBC cervical Benchmark dataset (4-classes) reveals a remarkable overall diagnostic accuracy of 97.94%. Furthermore, we employ visualization techniques to highlight infected areas in test images, while ROC plots further assess the performance of proposed model. Additionally, statistical validation through the McNemar test supports the robustness of the model. These results demonstrate that Cervi-Net is a reliable and effective tool for assisting medical professionals in the early and accurate diagnosis of cervical cancer.