Prior Guided Segmentation and Nuclei Feature Based Abnormality Detection in Cervical Cells

Abstract

Computer-assisted techniques for cytological analysis and abnormality detection, can help to early diagnose anomalies in cervical smear images. Cell nuclei carry substantial evidence of pre-cancerous changes, thus morphological properties of nuclei are important for accurate diagnostic decision. A novel nucleus feature-based cervical cell classification framework is proposed in this study. Prior guided segmentation algorithms are employed to accurately detect and segment nucleus. Fuzzy entropy based feature selection technique is used to select most discriminatory features, extracted from segmented nucleus. Five classifiers: k-nearest neighbor (KNN), linear discriminant analysis (LDA), Ensemble, and support vector machine with linear kernel (SVM-linear) and radial basis function kernel (SVM-RBF), are used to detect abnormality in cervical cells. The proposed framework is evaluated using Herlev dataset of 917 cervical cell images and compared with state-of-the-art methods. Results indicate that the proposed framework matches the performance of recent techniques, while segmenting nucleus and classifying Pap smear images using only 10 nucleus features. Therefore, the proposed abnormality detection framework can assist cytologists in computerized cervical cell analysis, and help with early discovery of any anomaly that may lead to cervical cancer.