An Explainable Multi-Level Framework for Cervical Cancer Detection Using Traditional Computer Vision and Deep Learning

Project Code :TCMAPY2420

Objective

The main objective of this project is to develop an automated cervical cancer detection system that utilizes deep learning models to classify cervical images into two categories: "cervical negative" and "cervical positive." By focusing on a deep learning-based approach, the project seeks to build a highly accurate system capable of diagnosing cervical cancer using cytology images. The system will incorporate a variety of pre-trained models, including CervixNet, MobileNetV2, and DenseNet121, to ensure the model’s robustness and scalability. The objectives of this project are to preprocess the dataset effectively, augment the images to improve model generalization, and evaluate the performance of different deep learning models using performance metrics such as accuracy, precision, recall, and F1-score. Furthermore, the project will explore the application of cross-validation to assess the models' robustness and ability to generalize across different subsets of the dataset. The ultimate aim is to develop a clinically useful tool that can assist healthcare professionals in early-stage detection of cervical cancer, providing a faster, more reliable alternative to manual image evaluation.

Abstract

Cervical cancer remains one of the leading causes of mortality among women worldwide. Early detection and classification of precancerous cells are critical for improving survival rates. In this work, we propose a deep learning-based framework for the automated detection of cervical cancer using the MendeleyLBC dataset. The proposed system leverages advanced pre-trained models, including EfficientNetB3, MobileNetV2, and DenseNet121, to classify cervical images into "cervical negative" and "cervical positive" categories. Each model is fine-tuned on the dataset to optimize performance and ensure accurate classification. The framework incorporates data preprocessing techniques such as image augmentation and normalization, followed by model training with five-fold cross-validation to ensure robustness. The models' performance is evaluated using metrics like accuracy, precision, recall, and F1-score. Our results show that the deep learning models, particularly DenseNet121 and EfficientNetB3, achieve high classification accuracy, with the potential to assist clinicians in early-stage cervical cancer diagnosis. Although the focus of this study is not on explain ability, future work could include integrating techniques like Grad-CAM or Integrated Gradients to provide insights into model decisions, making the system more interpretable for clinical use. The proposed approach paves the way for the development of accurate and clinically deployable automated systems for cervical cancer detection.

Keywords: Cervical cancer detection, deep learning, EfficientNetB3, MobileNetV2, DenseNet121,Mendeley LBC dataset, image classification.

NOTE: Without the concern of our team, please don't submit to the college. This Abstract varies based on student requirements.