Segmentation of Overlapping Cells in Cervical Cytology Images: A Survey

Abstract

Pap smear testing is crucial for early diagnosis of cervical cancer, but cell overlapping poses a significant challenge to diagnostic accuracy, as improper processing of overlapping cells can lead to misclassification. While significant research efforts have been devoted to segmenting overlapping cells, there is an absence of thorough reviews covering existing studies. This survey represents the first comprehensive exploration of technologies aiming to segment overlapping cells in cervical cytology images. Initially, we collected over 100 relevant papers from various open-source databases using diverse keywords. Subsequently, we conducted a thorough analysis covering various aspects, including datasets, evaluation methods, and data augmentation techniques. We then categorized the applications into conventional machine learning and deep learning approaches, further subdividing both methods into three groups. We summarized articles that utilized conventional machine learning methods and compared the outcomes with those employing deep learning methods. Finally, we provide insights into current challenges and prospects in this critical domain.

Keywords

• Cervical cells
• overlapping
• segmentation
• machine learning
• survey