Automated Screening of Precancerous Cervical Cells Using Self-Supervised Learning
Author Information
Author(s): Chun Jaewoo, Yu Ando, Ko Seokhwan, Chong Gunoh, Park Jiyoung, Han Hyungsoo, Park Nora Jeeyoung, Cho Junghwan
Primary Institution: Kyungpook National University
Hypothesis
Can a distribution-augmented contrastive self-supervised learning approach improve the detection of abnormal cervical cells from cytological images?
Conclusion
The proposed method enhances screening accuracy and reduces the workload for cytopathologists, contributing to more efficient cervical cancer screening programs.
Supporting Evidence
- The proposed DA-SimCLR (Color) model outperformed other models in detecting abnormal cervical cells.
- Using kernel density estimation (KDE) helped in identifying abnormalities in cytological samples.
- The model showed robustness against color distribution shifts, which is common in clinical settings.
Takeaway
This study created a smart computer program that helps doctors find bad cells in cervical cancer tests more easily and quickly.
Methodology
The study used a distribution-augmented contrastive learning approach to train a model on normal cervical cell images and evaluated its performance on detecting abnormal cells.
Limitations
The study's validation data were relatively low compared to the amount of trained data, and the model's decision-making interpretation based on cellular characteristics was not emphasized.
Participant Demographics
The internal cohort included 100 normal and 50 abnormal cervical cytology slides.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website