Detecting Abnormal Cervical Cells Based on Segmentation of Overlapping Cells

Abstract

Segmentation of overlapping cells in biological images remains a challenging task due to the complexity introduced by occlusion and overlap, which often leads to decreased accuracy in automated medical diagnostics systems. This research focuses on improving the segmentation of overlapping cells using an advanced deep learning approach. The model is trained on the ISBI2014 dataset of cervical cytology cells, a collection of images with noisy and overlapping cells. Experimental results demonstrate that the proposed model achieves higher accuracy, even with limited labeled data. By calculating the nucleus-to-cytoplasm (N/C) ratio from the segmentation results, the proposed method determines whether a cell is normal or exhibits pathological characteristics. The proposed method provides an efficient and scalable solution for the classification and segmentation of overlapping cells, offering new opportunities for advancements in automated image analysis and contributing to the development of more accurate and reliable systems for cell-based medical diagnosis.