Scholarlink

<p>The significance of early detection of <i>cervical</i> abnormalities cannot be overstated due to the severe threat posed by <i>cervical cancer</i> globally. However, the current utilization of Pap smear and Cytology-based screening methods is limited by their high cost and labour-intensive nature, limiting accessibility for many individuals. To address this issue a fully automated computer-aided framework named <i>e-TransCoder</i> utilizing <i>Deep Learning (DL)</i> for cytology image classification is proposed. By leveraging sophisticated techniques such as <i>EfficientNetB0</i> and <i>Vision Transformer</i> (ViT), along with a dimension reduction approach, <i>Autoencoder</i>, our framework aims to enhance diagnostic accuracy and optimize screening processes. This approach is applied across two different publicly available datasets named <i>SipakMed</i> and <i>Herlev</i>, consisting of single-cell <i>Pap-Smear</i> images, with one dataset serving for training and validation and the other for testing. The hybrid deep feature engineering network, along with data augmentation, exhibits robust performance. Various performance metrics, including <i>Accuracy, Matthews Correlation Coefficient, Receiver Operating Characteristic - Area Under the Curve (ROC-AUC) Score, Cohen’s Jaccard Coefficient, Sensitivity, and Specificity</i>, are employed to assess the model. For binary-class classification, the proposed approach achieves an <i>accuracy</i> of <i>98.26%</i>, <i>Matthews Correlation Coefficient (MCC)</i> of <i>95.54%</i>, <i>Cohen’s Jaccard Coefficient (Cohen’s J)</i> of <i>93.65%</i>, <i>sensitivity</i> of <i>97.52%</i>, <i>specificity</i> of <i>98.52%</i>, and an <i>AUC</i> score of <i>99%</i>. Despite the discrepancy in dataset dimensions, the findings of this study highlight the importance of leveraging <i>DL</i> techniques in medical image analysis. Moreover, the development of a user-friendly <i>Graphical User Interface (GUI)</i> framework may facilitate physician intervention, potentially enhancing the overall efficacy of early prediction of <i>cervical cancer</i> and ultimately improving the treatment of affected patients.</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

Detecting cervical cancer from pap smear images using e-transcoder based deep learning technique

  • 期刊论文

  • 发表日期：

    2026年4月22日

查看全文
  • Poonam Moral,
  • Debjani Mustafi,
  • Abhijit Mustafi,
  • Sudip Kumar Sahana

摘要

The significance of early detection of cervical abnormalities cannot be overstated due to the severe threat posed by cervical cancer globally. However, the current utilization of Pap smear and Cytology-based screening methods is limited by their high cost and labour-intensive nature, limiting accessibility for many individuals. To address this issue a fully automated computer-aided framework named e-TransCoder utilizing Deep Learning (DL) for cytology image classification is proposed. By leveraging sophisticated techniques such as EfficientNetB0 and Vision Transformer (ViT), along with a dimension reduction approach, Autoencoder, our framework aims to enhance diagnostic accuracy and optimize screening processes. This approach is applied across two different publicly available datasets named SipakMed and Herlev, consisting of single-cell Pap-Smear images, with one dataset serving for training and validation and the other for testing. The hybrid deep feature engineering network, along with data augmentation, exhibits robust performance. Various performance metrics, including Accuracy, Matthews Correlation Coefficient, Receiver Operating Characteristic - Area Under the Curve (ROC-AUC) Score, Cohen’s Jaccard Coefficient, Sensitivity, and Specificity, are employed to assess the model. For binary-class classification, the proposed approach achieves an accuracy of 98.26%, Matthews Correlation Coefficient (MCC) of 95.54%, Cohen’s Jaccard Coefficient (Cohen’s J) of 93.65%, sensitivity of 97.52%, specificity of 98.52%, and an AUC score of 99%. Despite the discrepancy in dataset dimensions, the findings of this study highlight the importance of leveraging DL techniques in medical image analysis. Moreover, the development of a user-friendly Graphical User Interface (GUI) framework may facilitate physician intervention, potentially enhancing the overall efficacy of early prediction of cervical cancer and ultimately improving the treatment of affected patients.