HPV genotype-specific p16/Ki67 expression with machine learning-assisted assessment in cervical neoplasia
Frontiers in Oncology, cilt.16, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 16
- Basım Tarihi: 2026
- Doi Numarası: 10.3389/fonc.2026.1819567
- Dergi Adı: Frontiers in Oncology
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, EMBASE, Directory of Open Access Journals
- Anahtar Kelimeler: artificial intelligence, cervical intraepithelial neoplasia, digital pathology, HPV genotyping, machine learning, p16/Ki67 biomarkers
- Sağlık Bilimleri Üniversitesi Adresli: Evet
Özet
Introduction – Genotype-specific patterns of dual p16/Ki67 immunoexpression and their integration with computational assessment remain insufficiently delineated in cervical neoplasia. The present investigation characterized biomarker expression stratified by HPV genotype and evaluated the methodological feasibility of a deep learning–assisted scoring pipeline. Methods – A single-center cross-sectional investigation was conducted on 100 HPV-positive women, stratified into three categories: HPV16 mono-infection (n=33), non-HPV16 high-risk mono-infection (n=33), and multi-genotype co-infection (n=34). Whole-slide p16/Ki67 immunohistochemistry was scored through real-time consensus by two pathologists of differing experience levels, blinded to HPV genotype. A ResNet50-based computational pipeline was developed as a methodological feasibility demonstration and evaluated on an independent held-out test set (n=25). Between-group comparisons were performed using the Mann–Whitney U test with Bonferroni correction; multivariable logistic regression and receiver operating characteristic (ROC) analysis were employed to quantify biomarker discriminative performance; and analysis of covariance (ANCOVA) with age as a continuous covariate was undertaken as a pre-specified sensitivity analysis. Results – HPV16 mono-infection exhibited significantly elevated p16 immunoexpression (52.4 ± 27.6%) relative to non-HPV16 high-risk genotypes (31.1 ± 22.8%; p=0.0021; Cohen’s d=0.844) and co-infections (34.6 ± 24.1%); between-group differences persisted following age adjustment. On the independent test set, the computational model yielded an accuracy of 96.0% (95% CI, 78.3–99.9%) and an AUC-ROC of 0.96. Given the restricted test set dimension and single-center design, these metrics should be construed as preliminary. Conclusions – HPV16 mono-infection is associated with distinctly elevated dual p16/Ki67 immunoexpression, providing methodological support for genotype-informed cytological risk stratification. The computational pipeline demonstrates technical feasibility; however, external multicenter validation is required prior to any consideration of clinical implementation.