A Hybrid Approach for Cervical Cancer Classification: Combining D-CNNs, Transfer Learning, and Ensemble Models

Abstract

Cervical cancer is the top cause of cancer deaths in women globally, despite being easily preventable through early diagnosis and treatment. This papers suggested deep learning methods, particularly transfer learnings, deep convolutional neural network (D-CNN), and ensemble learnings for automating the detection and classification of cervical cancer. In particular, we assess the effectiveness of four deep convolutional neural network structures: AlexNet, ZfNet, HighwayNet, and LeNet-5, in addition to four transfer learning architectures: EfficientNetB1, ResNet151, MobileNetV3, and DenseNet211. The dataset underwent preprocessing from the start. To achieve this, we conducted error level analysis (ELA) on the dataset to confirm that no patterns were overlooked within each image. We additionally carried out augmentation on the dataset (resizing, rescaling, flipping, rotating, zooming, and contrasting). Enhanced diagnostic accuracy can be attained through deep learning applied to the multi-cancer dataset. Comparative analyses were carried out swiftly to explore the precision of these architectures. We introduce a new hybrid ensemble model, AZL, which integrates AlexNet, ZfNet, and LeNet-5 to address the limitations of each individual model based on performance comparisons. We evaluated all these models in an experimental setup. The results of our experiments indicate that the AZL ensemble model reached a classification accuracy of 99.92%, surpassing individual D- CNNs and transfer learning models in precision, recall, and F1-score. These results emphasize the efficacy of ensemble deep learning methods in enhancing cervical cancer diagnosis. The method we developed shows potential to assist pathologists in diagnosing this disease promptly, particularly considering the limited resources available. In the end, it can precisely.