Multi-Class Segmentation with Deep Learning based Pap Smear Image Analysis for Cervical Cancer Detection and Classification Model

Abstract

Cervical cancer (CC) is a significant global health issue, but earlier diagnosis and accurate detection could considerably enhance treatment outcomes and reduce mortality rates. Pap smear test is a popular screening technique for CC. But the manual interpretation of Pap smear images can be subjective and error-prone. Deep learning (DL) technique assists to automate and enhance the performance of CC detection on Pap smear images. DL model continuously enhances with more diverse and larger datasets, potentially resulting in even more reliable and more accurate diagnoses. In this aspect, this study introduces an Automated Deep Learning based Pap Smear Image Analysis for Cervical Cancer Detection and Classification (ADLPSIA-CCD) method. The objective of the ADLPSIA-CCD technique is to utilize DL models for the segmentation and classification of Pap smear images for CC diagnosis. In the presented ADLPSIA-CCD method, the initial stage of image preprocessing takes place in two levels such as image sharpening using Gaussian blur and contrast enhancement. In addition, the ADLPSIA-CCD technique exploits U-Net based image segmentation process with EfficientNetB3 as a backbone network. Next, radiomics feature extraction process takes place where the features in every segmented class are extracted separately. Finally,Conv-Recurrent Hopfield Neural Network (CRHNN) model is exploited for the classification of CC. Notably, the CRHNN model trains the segmented class features separately. The simulation outcome of the ADLPSIA-CCD algorithm is tested against a benchmark dataset. The comprehensive result analysis showed the superior performance of the ADLPSIA-CCD technique on CC detection.