Scholarlink

In recent years, ML and CV techniques have made significant contributions to medical diagnostics. This paper explores the application of Transfer Learning for classifying cervical cancer images from Pap smear and colposcopy tests. The objective is to find the best transfer learning models for classification of 7 cervical lesion categories of 542 images from Malhari Dataset for early cancer detection using Binary classification and Multiclass Classification. The proposed model achieves significant performance in terms Metrics like precision, accuracy and recall etc. compared to traditional models. Our results demonstrate the potential for MobileNet in Binary Classification providing 100% accuracy and DenseNet201 providing 99.34% accuracy in Multiclass Classification with 99.34% recall and other transfer learning techniques like VGG16, ResNet50, InceptionV3, Xception, EfficientNetB0 etc. along with DCGAN. Further pre-trained Models were used to deploy explainable AI to reveal biomarkers. Of these, the best two models were picked up to provide for a third model which provided us with maximum accuracy and recall. In this work, we introduce InDraNet, a hybrid deep learning model that combines the strengths of InceptionV3 and DenseNet201 for robust image classification. Through the strategic use of pre-trained models, feature fusion, and layer freezing, the architecture offers a powerful way of classifying wide variety of images. Overall motivation behind this research is to reduce the challenge of false negative cases in medical diagnosis of cervical cancer which lead to diagnosis in the later stages at which treatment becomes challenging.

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

InDraNet: A Novel Explainable Framework for Cervical Cancer Risk Assessment Leveraging Transfer Learning

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Nayan Saxena,
  • Shailesh D. Kamble

摘要

In recent years, ML and CV techniques have made significant contributions to medical diagnostics. This paper explores the application of Transfer Learning for classifying cervical cancer images from Pap smear and colposcopy tests. The objective is to find the best transfer learning models for classification of 7 cervical lesion categories of 542 images from Malhari Dataset for early cancer detection using Binary classification and Multiclass Classification. The proposed model achieves significant performance in terms Metrics like precision, accuracy and recall etc. compared to traditional models. Our results demonstrate the potential for MobileNet in Binary Classification providing 100% accuracy and DenseNet201 providing 99.34% accuracy in Multiclass Classification with 99.34% recall and other transfer learning techniques like VGG16, ResNet50, InceptionV3, Xception, EfficientNetB0 etc. along with DCGAN. Further pre-trained Models were used to deploy explainable AI to reveal biomarkers. Of these, the best two models were picked up to provide for a third model which provided us with maximum accuracy and recall. In this work, we introduce InDraNet, a hybrid deep learning model that combines the strengths of InceptionV3 and DenseNet201 for robust image classification. Through the strategic use of pre-trained models, feature fusion, and layer freezing, the architecture offers a powerful way of classifying wide variety of images. Overall motivation behind this research is to reduce the challenge of false negative cases in medical diagnosis of cervical cancer which lead to diagnosis in the later stages at which treatment becomes challenging.