Evaluation of the Diagnostic Accuracy of Cervical Cell Morphologies from Android Device-Captured Cytopathological Microscopic Images through Artificial Intelligence in Mainly Rural or Resource-Constraint Areas of India

Document Type : Research Articles

Authors

1 Department of Information Technology, Cell and Molecular Biology, CliniMed LifeSciences, Kolkata, West Bengal, India.

2 Department of Pathology, Calcutta National Medical College and Hospital, Kolkata, West Bengal, India.

3 Department of Pathology, Nil Ratan Sircar Medical College and Hospital, Kolkata, West Bengal, India.

4 Department of Oncology, Chittaranjan National Cancer Institute, Kolkata, West Bengal, India.

Abstract

Objective: This study aims at develop and evaluate an artificial intelligence programming software, an integrated system that automatically detects and classifies cells from microscopic Pap smear slide images taken on Android phones or tabs to diagnose the cervical cell morphology in a time-efficient and cost-effective manner. Methods: This study presents an integrated system designed to automatically detect and classify cells in Pap smear slide images, differentiating cellular morphologies. The system leverages three deep learning (DL) and one machine learning (ML) models, each tailored to specific tasks in the image analysis pipeline. The analysis of 292 hospital in-house microscopic Pap smear images was conducted from July 2023 to December 2024 at CliniMed LifeSciences, Kolkata, India. The following article describes the datasets used, the training procedures and the performance metrics for each model.  Results: Pap smear images have been validated and standardized by using SipakMed, Herlev (public datasets) and hospital in-house data. A total of 292 in-house Pap smear images have been analysed through the newly developed AI software. Standardization and validation include an Intersection-over-Union score of cell-nuclei boundary extraction model of 71.14%, the accuracy of cell classification model and morphological feature based ML model are 99.213% and 91.23% respectively. The custom AI model could successfully classify 98.09% and 80.49%  of normal and abnormal cells in hospital in-house samples respectively. Also a significant meaningful correlation is observed between biopsy (gold standard) and AI reports. Conclusion: AI offers a lot of promise for diagnosing cervical cancer, and its uses in cervical cytology screening are particularly well-established. Manual screening of cervical cytology smears is a time-tested method, but AI is set to revolutionize the process by improving outreach, availability, accuracy and economy. A total of 292 hospital in-house Pap smear images have been validated and examined in this study with significant accuracy percentages between AI and expert eyes. 

Keywords

Main Subjects

Asian Pacific Journal of Cancer Prevention
Volume 27, Issue 2
February 2026
Pages 601-612

Files

History

  • Receive Date: 07 July 2025
  • Revise Date: 27 September 2025
  • Accept Date: 04 February 2026

Share

How to cite

Statistics