A comparative study of deep learning architectures for cervical cell segmentation in Pap smear images

Name: Palaparthi, Revanth

Organization : IIT Dharwad. Dept. of CSE

Email : 210010043@iitdh.ac.in

Name: Chinta, Sahitya

Organization : IIT Dharwad. Dept. of CSE

Email : 210010045@iitdh.ac.in

Name: Samad, Shaik Abdul

Organization : IIT Dharwad. Dept. of EECE

Email : ee23mt018@iitdh.ac.in

Name: Pandya, Rahul J.

Organization : IIT Dharwad. Dept. of EECE

Email : rpandya@iitdh.ac.in

LCSH: Cervix uteri -- Cancer -- Cytodiagnosis

LCSH: Image segmentation

LCSH: Deep learning (Machine learning)

LCSH: Neural networks (Computer science)

LCSH: Diagnostic imaging

Abstract: Segmentation of cervical cells is an essential step in the early detection and diagnosis of cervical cancer. This study proposes a comprehensive deep learning pipeline for binary segmentation of cervical cells in Pap smear images based on eight advanced and recent architectures: U-Net, Attention U-Net, U-Net++, DeepLabv3+, PAN, SegFormer, YOLOv8, and a self-designed YOLOv11. The models were trained and tested using the publicly available SIPaKMeD dataset, which contains 966 high-resolution labeled cytology images. All images were divided into overlapping 512 × 512 patches, generating over 8,000 training patches and 1,208 testing patches. Data augmentation techniques—including flipping, rotation, elastic deformation, and contrast adjustment—were applied to increase variability. Segmentation performance was quantitatively evaluated using six metrics: Dice coefficient, Intersection over Union (IoU), precision, recall, F1-score, and accuracy. The U-Net++ model achieved the highest performance with a Dice score of 0.8366 and an IoU score of 0.7771. The YOLOv11 model demonstrated competitive performance with a Dice score of 0.8349 and an IoU of 0.7453, while providing reduced inference time. These findings highlight the trade-offs between segmentation accuracy and computational speed, and indicate that the proposed pipeline has potential for integration into real-time clinical diagnostic systems.

King Mongkut's University of Technology North Bangkok. Central Library

Address: BANGKOK

Email: library@kmutnb.ac.th

Created: 2025

Modified: 2026-02-24

Issued: 2026-02-24

บทความ/Article

application/pdf

BibliograpyCitation : In Prince of Songkla University, Phuket Campus. College of Computing. The 9th International Conference on Information Technology (InCIT 2025) (pp.176-182). Phuket : Prince of Songkla University

eng

©copyrights King Mongkut's University of Technology North Bangkok

ใช้เวลา

0.029584 วินาที