Cervical Cell Segmentation and Classification Using U-Net and Hybrid VGG19-AlexNet Architecture

Cervical cancer develops when some cells in the cervix become malignant. Symptoms do not appear during the precancerous stage. This study aims to segment images into whole cells and backgrounds to identify the region of interest (ROI) in detecting cervical cancer. Segmentation is crucial for obtaining features in cervical cell images, such as the cytoplasm. The segmentation output multiplied by the real images serves as the input for the classification task, which classifies cervical cell types. The U-Net architecture is used for the segmentation task, and VGG19 with AlexNet is added as dense layers for classifying cervical cells. The Intersection over Union (IoU) and accuracy of the segmentation result achieved 99.98% and 98.3%, respectively. For the classification task, the proposed model combining VGG19 and AlexNet achieved an accuracy of 93%, outperforming the individual VGG19 and AlexNet models, which scored 92% and 85%, respectively, in classifying cervical cells. In conclusion, the results of this study are promising for the detection of cervical cancer.