Deep learning-based automatic segmentation and classification for cervical cancer detection using an improved U-Net and ensemble methods

Cervical cancer is the fourth leading cause of cancer-related illness and death among women worldwide. The Pap test is a widely used screening technique to detect abnormal cells that can become cancerous. In this research, we proposed a method for automatic segmentation and classification of cervical cancer cell images. The method uses an improved U-Net architecture to segment the image and identify the region of interest (ROI). Following segmentation, we classify the cervical cell type using ResNet50V2 and an ensemble of different pretrained models to enhance performance. We developed several pipelines for cervical cancer detection, including a normal method, with and without RES_DCGAN, before classification and segmentation tasks. The proposed method was evaluated using the Pomeranian, Herlev, and SIPaKMeD datasets. The experimental results showed that whole-cell segmentation achieved 99.53%, 88.95%, and 98.3% accuracy when RES_DCGAN was added before the segmentation. The framework achieved 96% and 95% accuracy for multi-class classification on the Pomeranian and SIPaKMeD datasets, respectively. Additionally, the Herlev dataset scored an accuracy of 91%, while SIPaKMeD achieved 99% for the binary classification of cervical cell types using the ensemble method. In conclusion, the deep learning-based segmentation and classification method demonstrated promising results for cervical cancer detection and can help pathologists diagnose the disease.