Self-supervised learning-based cervical cytology for the triage of HPV-positive women in resource-limited settings and low-data regime.

Screening Papanicolaou test samples has proven to be highly effective in reducing cervical cancer-related mortality. However, the lack of trained cytopathologists hinders its widespread implementation in low-resource settings. Deep learning-assisted telecytology diagnosis emerges as an appealing alternative, but it requires the collection of large annotated training datasets, which is costly and time-consuming. In this paper, we demonstrate that the abundance of unlabeled images that can be extracted from Pap smear test whole slide images presents a fertile ground for self-supervised learning methods, yielding performance improvements compared to off-the-shelf pre-trained models for various downstream tasks. In particular, we propose Cervical Cell Copy-Pasting (C3P) as an effective augmentation method, which enables knowledge transfer from public and labeled single-cell datasets to unlabeled tiles. Not only does C3P outperforms naive transfer from single-cell images, but we also demonstrate its advantageous integration into multiple instance learning methods. Importantly, all our experiments are conducted on our introduced in-house dataset comprising liquid-based cytology Pap smear images obtained using low-cost technologies. This aligns with our long-term objective of deep learning-assisted telecytology for diagnosis in low-resource settings.

Telecytologic diagnosis of cervical smears for triage of self-sampled human papillomavirus-positive women in a resource-limited setting: concept development before implementation.

INTRODUCTION: Cytology is an option for triaging human papillomavirus (HPV)-positive women. The interpretation of cytologic slides requires expertise and financial resources that are not always available in resource-limited settings. A solution could be offered by manual preparation and digitization of slides on site for real-time remote cytologic diagnosis by specialists. In the present study, we evaluated the operational feasibility and cost of manual preparation and digitization of thin-layer slides and the diagnostic accuracy of screening with virtual microscopy. MATERIALS AND METHODS: Operational feasibility was evaluated on 30 cervical samples obtained during colposcopy. The simplicity of the process and cellularity and quality of digitized thin-layer slides were evaluated. The diagnostic accuracy of digital versus glass slides to detect cervical intraepithelial neoplasia grade 2 or worse was assessed using a cohort of 264 HPV-positive Cameroonian women aged 30 to 49 years. The histologic results served as the reference standard. RESULTS: Manual preparation was found to be feasible and economically viable. The quality characteristics of the digital slides were satisfactory, and the mean cellularity was 6078 squamous cells per slide. When using the atypical squamous cells of undetermined significance or worse threshold for positivity, the diagnostic performance of screening digital slides was not significantly different statistically compared with the same set of slides screened using a light microscope (P = 0.26). CONCLUSIONS: We have developed an innovative triage concept for HPV-positive women. A quality-ensured telecytologic diagnosis could be an effective solution in areas with a shortage of specialists, applying a same day "test-triage-treat" approach. Our results warrant further on-site clinical validation in a large prospective screening trial.