Classification of Multiclass Pap Smear Image and Asymmetric Characterization for Prediction of Severity of Cervical Abnormality

Abstract

Cervical cancer is the fourth most common cancer among women globally. This study introduces a computer-aided hybrid framework to enhance cervical abnormality prediction based on Pap smear cytological images, facilitating efficient screening and early detection. Given the known morphological and textural differences between the nucleus of a normal cervical cell and the nucleus of a precancerous cervical cell, our proposed framework consists of three phases. First, the nucleus regions in Pap smear images are segmented. Next, both shape and texture features are extracted from the segmented nucleus and the holistic images, respectively. Finally, a support vector machine (SVM) is employed to classify the images into two, three, and four classes using the extracted features. We achieved classification accuracy, specificity, sensitivity, F1 score, and G-measure values of 89.13%, 89.67%, 87.59%, 88.11%, and 87.48%, respectively, for shape feature–based classification. For the combined shape and texture feature–based classification, these metrics were 85.25%, 84.78%, 84.64%, 85.26%, and 85.92%, respectively. Notably, the classification based on shape features outperformed both texture feature–based and the combination of shape and texture feature–based classifications. Additionally, the performance of 14 state-of-the-art deep learning models was evaluated, with shape features delivering superior accuracy compared to deep learning methods. Overall, we conclude that nucleus-based shape features of normal and abnormal cell images, along with the proposed method, have strong potential for automating cervical abnormality detection.