Abstract

Cervical cancer is the fourth most common cancer among women worldwide, with a disproportionate burden on low- and middle-income countries (LMICs), particularly in Sub-Saharan Africa, where Nigeria records the highest incidence and mortality. Limited access to specialist colposcopists and the subjectivity of manual colposcopy make prevention-oriented, automated triage tools critically needed. Existing deep learning models largely focus on distinguishing advanced cancer from normal tissue rather than detecting the three precancerous stages, the clinically critical window for intervention. This study investigates two complementary self-supervised and domain-adaptive deep learning approaches for cervical cancer triage from colposcopy images, evaluated on a patient-level dataset of 309 patients from Bowen University, Nigeria. Method 1 employs a Vision Transformer-Base/16 (ViT-B/16) encoder pretrained using masked autoencoder (MAE) self-supervised learning on an open-source colposcopy dataset spanning the three precancerous stages and compares it against three baselines (ImageNet ViT-B/16, EndoViT, and EfficientNet-B2) using identical fine-tuning. A preliminary single-split evaluation showed that the domain-adaptive MAE encoder achieved a precancer AUC of approximately 0.75 and a sensitivity of approximately 0.92, compared to approximately 0.68 for ImageNet ViT-B/16 and approximately 0.65 for EndoViT. Method 2 implements a SimCLR contrastive self-supervised pretraining pipeline using an EfficientNet-B3 backbone, leveraging unlabelled images from both datasets to learn cervical visual representations before supervised fine-tuning for binary Normal/Abnormal classification, achieving an accuracy of 0.8, recall of 0.71, and AUC of 0.83 on the held-out test set.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MIRASOL_052.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to Open Review

Open Review Page: https://openreview.net/forum?id=WkJybXtxEQ

BibTex

@InProceedings{VicAye_PreventionOriented_MICCAISAT2026,
        author = { Vicky, Ayebare AND Andrew, Mudhasi AND Namuhubembe, Melisa Grace AND Matovu, Steven Ziritwawula AND Williams, Mayinja Bob AND Raymond, Confidence AND Emegoakor, Adaobi Chiazor AND Bwire, Patience AND Iorumbur, Aondona Moses},
        title = { { Prevention-Oriented Cervical Cancer Triage Using a Nigerian AI Screening Dataset } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 17265},
        month = {pending},
        page = {pending}
}


back to top