EUVIMEDEUVIMED entdecken
Crossref

Deep learning-assisted digital VIA via telemedicine and HPV self-sampling for CIN2+ detection in remote archipelago women: a prospective diagnostic accuracy study

Theresia Putri Sinaga, Nur Diana, Firman Hadi, Gayatri Putri

Sriwijaya Journal of Obstetrics and Gynecology · 21.07.2026 · 4 (1) · 25-31

DOI: 10.59345/sjog.v4i1.302 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Sriwijaya Journal of Obstetrics and Gynecology, online erschienen am 21.07.2026. Verfasst von Theresia Putri Sinaga und 3 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Background

Cervical cancer disproportionately burdens women in low- and middle-income countries, where archipelagic geography and colposcopist shortages obstruct screening. Human papillomavirus (HPV) self-sampling is highly sensitive but poorly specific, generating colposcopy referrals that exceed remote-area capacity.

Objective

To evaluate whether deep-learning (DL)-assisted digital visual inspection with acetic acid (VIA), delivered by telemedicine, can triage HPV-positive women and detect high-grade cervical intraepithelial neoplasia (CIN2+).

Methods

In a prospective, double-blind, STARD-compliant diagnostic accuracy study, 642 women aged 30–50 years at an urban tertiary referral hospital (Center A) and five remote community health centers (Region B) in an Indonesian archipelago province underwent HPV-DNA self-sampling and smartphone-captured digital VIA analyzed by a MobileNetV2 convolutional neural network. All participants received colposcopy-directed biopsy as the reference standard, eliminating verification bias. Metrics used Wilson 95% confidence intervals (CI); multivariable logistic regression and decision-analytic triage metrics were derived.

Results

CIN2+ prevalence was 13.1% (95% CI 10.7–15.9). DL-assisted VIA achieved sensitivity 91.7% (95% CI 83.8–95.9), specificity 88.4% (85.4–90.8), and AUC 0.93, exceeding human-read VIA (sensitivity 67.9%). HPV self-sampling was most sensitive (95.2%) but least specific (81.5%). A sequential HPV→DL-VIA pathway raised specificity to 95.7% and positive predictive value to 75.5%, reducing colposcopy referrals by 46.4% and unnecessary referrals by 76.7% (number-needed-to-screen 7.6). HPV positivity dominated the multivariable model (adjusted OR 91.5, 95% CI 32.7–256.2, p

Zitieren

APA

Sinaga, T. P., Diana, N., Hadi, F., & Putri, G. (2026). Deep learning-assisted digital VIA via telemedicine and HPV self-sampling for CIN2+ detection in remote archipelago women: a prospective diagnostic accuracy study. Sriwijaya Journal of Obstetrics and Gynecology, 4(1), 25-31. https://doi.org/10.59345/sjog.v4i1.302

Vancouver

Sinaga TP, Diana N, Hadi F, Putri G. Deep learning-assisted digital VIA via telemedicine and HPV self-sampling for CIN2+ detection in remote archipelago women: a prospective diagnostic accuracy study. Sriwijaya Journal of Obstetrics and Gynecology. 2026;4(1):25-31. doi:10.59345/sjog.v4i1.302

BibTeX
@article{Sinaga2026,
  author = {Sinaga, Theresia Putri and Diana, Nur and Hadi, Firman and Putri, Gayatri},
  title = {Deep learning-assisted digital VIA via telemedicine and HPV self-sampling for CIN2+ detection in remote archipelago women: a prospective diagnostic accuracy study},
  journal = {Sriwijaya Journal of Obstetrics and Gynecology},
  year = {2026},
  volume = {4},
  number = {1},
  pages = {25--31},
  doi = {10.59345/sjog.v4i1.302}
}
Kostenlos starten →