AUTOMATED CERVICAL CELL NUCLEI SEGMENTATION BASED ON MULTILAYER UNSUPERVISED CLUSTERING ALGORITHM AND MORPHOLOGICAL APPROACH

• Main Authors: Khalis Danial Nukman, Khiruddin, Wan Azani, Mustafa, Khairur Rijal, Jamaludin, Khairul Shakir, Ab Rahman, Hiam, Alquran, Syahrul Nizam, Junaini
• Format: Article
• Language: English
• Published: HOEHERE BUNDESLEHRANSTALT UND BUNDESAMT FUER WEIN- UND OBSTBAU 2025
• Subjects: QA75 Electronic computers. Computer science
• Online Access: http://ir.unimas.my/id/eprint/47661/1/AUTOMATED%20CERVICAL.pdf; http://ir.unimas.my/id/eprint/47661/; https://www.researchgate.net/publication/388959013_AUTOMATED_CERVICAL_CELL_NUCLEI_SEGMENTATION_BASED_ON_MULTILAYER_UNSUPERVISED_CLUSTERING_ALGORITHM_AND_MORPHOLOGICAL_APPROACH
• Institution: Universiti Malaysia Sarawak

Cervical cancer, a leading cause of female mortality globally, results from abnormal cell growth in the cervix, making early detection crucial. This study suggests an automated segmentation approach that is more accurate and faster than traditional methods, which face challenges such as contrast problems and noise. The research aims to develop an algorithm for autonomously segmenting the nucleus of cervical cells to aid in diagnosis and future research. The proposed methodology involves extracting and enhancing the brightness (V channel) of input images using a median filter and Pairing Adaptive Gamma Correction and Histogram Equalisation (PAGCHE). A segmentation method based on multiple Fuzzy C-Means Clustering (FCM) layers and flexible morphological approaches is used to segment the nuclei in Pap smear images. The study utilized 917 images from the Herlev dataset to evaluate the method's performance. Image Quality Assessment (IQA) metrics, including accuracy, sensitivity, precision, specificity, and F-measure, demonstrate the method's efficacy. Results show the proposed approach consistently achieves over 90% accuracy. It outperforms other methods like Chan-Vese (CV), Canny edge-based, adaptive threshold, and FCM, with the highest accuracy, F1-measure, and sensitivity at 92.19%, 94.40%, and 93.38%, respectively. It also ranks second in precision and specificity, at 96.41% and 94.25%. These results indicate the approach's high accuracy, sensitivity, and specificity, making it a reliable tool for early detection and diagnosis. The algorithm's successful implementation could improve patient outcomes and support further research in cervical cancer diagnostics. The average segmentation score of the 917 images exceeds 90%, highlighting the method's flexibility.