Related Experiment Video
Updated: Jul 9, 2025
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
1.9K
Reproducible and clinically translatable deep neural networks for cervical screening
Syed Rakin Ahmed1,2,3,4, Brian Befano5,6, Andreanne Lemay7,8
1Athinoula A. Martinos Center for Biomedical Imaging, Department of Radiology, Massachusetts General Hospital, Boston, MA, 02129, USA. syedrakin_ahmed@fas.harvard.edu.
Scientific Reports
|December 8, 2023
Summary
Artificial intelligence (AI) can improve cervical cancer screening in low-resource settings. A deep learning model achieved high accuracy in detecting precancerous lesions from cervical images, aiding human papillomavirus (HPV) positive triage.
Area of Science:
- Oncology
- Medical Imaging
- Artificial Intelligence
Background:
- Cervical cancer is a major cause of mortality, disproportionately affecting low- and middle-income countries (LMICs).
- Current cervical screening methods face challenges in scalability and cost-effectiveness in LMICs.
- Visual triage of human papillomavirus (HPV)-positive individuals is recommended by the WHO for secondary prevention.
Purpose of the Study:
- To develop and optimize a deep learning (AI) pipeline for accurate diagnosis of cervical precancerous lesions using digital images.
- To evaluate the performance, portability, and repeatability of AI models for cervical screening.
- To assess the potential of AI to assist in visual triage for women testing positive for HPV.
Main Methods:
- A comprehensive deep learning model selection and optimization study was conducted.
- A large, multi-geography, multi-institution, and multi-device dataset of 9,462 women (17,013 images) was utilized.
- Model performance was evaluated using metrics including Area Under the Receiver Operating Characteristics Curve (AUC), quadratic weighted kappa (QWK), and misclassification rates.
Main Results:
- The top-performing deep learning model, combined with HPV type, achieved an AUC of 0.89.
- The model demonstrated a low total extreme misclassification rate of 3.4% on held-aside test sets.
- High consistency and reliability were observed, with a QWK of 0.86 and minimal 2-class disagreement (0.69%).
Conclusions:
- This study presents one of the first robust, repeatable, and clinically translatable deep learning models for cervical screening.
- The developed AI model shows significant promise for improving the accuracy and efficiency of cervical cancer secondary prevention, particularly in resource-limited settings.
- AI-assisted cervical screening has the potential to overcome existing infrastructure and cost barriers in LMICs.
