A hybrid approach for cervical cancer detection: Combining D-CNN, transfer learning, and ensemble models

Abstract

Cervical cancer is the leading cause of cancer-related death among women worldwide, although it is easily preventable through early detection and treatment. This paper proposed deep learning techniques, specifically transfer learning, deep convolutional neural networks (D-CNNs), and ensemble learning for automating cervical cancer detection and classification. Specifically, we evaluate the performance of four deep convolutional neural networks architectures: AlexNet, ZfNet, HighwayNet, and LeNet-5, as well as four transfer learning architectures: EfficientNetB0, ResNet50, MobileNetV2, and DenseNet201. The dataset was preprocessed from the beginning. For this, we performed error level analysis (ELA) on the dataset to ensure that no patterns were missed within each image. We also performed augmentation on the dataset (resizing, rescaling, flipping, rotation, zooming, and contrasting). It is possible to achieve improved diagnostic accuracy using deep learning on the multi-cancer dataset. Comparative studies were conducted in a very short time to investigate the accuracy of these architectures. Based on performance comparisons, we propose a novel hybrid ensemble model AZL which combines AlexNet, ZfNet, and LeNet to overcome individual model limitations. We compared all these models in an experimental format. Our experimental results show that the AZL ensemble model achieved a classification accuracy of 99.92 %, outperforming individual D-CNN and transfer learning models in terms of precision, recall, and F1-score. These findings highlight the effectiveness of ensemble deep learning approaches in improving cervical cancer diagnosis. Our developed method holds promise to help pathologists diagnose this disease in a timely manner, especially given the limited resources. Ultimately, it is capable of accurately detecting and classifying cervical cancer, which contributes to reducing mortality.