Pap smear & AI: the new era of cervical screening

By Ana Espino | Published on February 26, 2026 | 3 min read

Cervical cancer remains one of the few largely preventable cancers thanks to organized screening programs combining cervical cytology and HPV testing. Yet, despite these established strategies, diagnoses at advanced stages still occur. Conventional cytology, the historical cornerstone of screening, has intrinsic limitations. Its interpretation relies on expert human morphological analysis, subject to inter- and intra-observer variability. Low-grade lesions, subtle atypia, and paucicellular samples represent diagnostic weak points. As screening programs expand, the growing volume of samples increases the risk of analytical fatigue and triage errors.  

In this already demanding context, the introduction of HPV testing has transformed decision-making algorithms. More sensitive for identifying at-risk patients, it enhances early detection. However, its lower specificity increases the number of follow-up examinations and complicates clinical stratification. The challenge is therefore no longer solely to detect abnormalities, but to accurately prioritize risk while maintaining laboratory efficiency.  

Amid the transition toward more standardized and digital medicine, artificial intelligence (AI) has emerged as a strategic lever. A review published in 2025 in Bioengineering examined the growing integration of AI systems in diagnostic cervical cytology, assessing their analytical performance, organizational potential, and clinical limitations.  

Can AI reduce false negatives?

The authors conducted a narrative review based on literature from the past five years. Thirty studies were included, primarily focusing on clinical applications of deep learning in cervical cytology. Bibliometric analysis revealed rapid expansion of the field, with 90% of publications occurring within the last five years, reflecting increasing scientific interest.  

Reported diagnostic performance was high. Several models based on convolutional neural networks (CNNs) or hybrid architectures achieved performance comparable to, or exceeding, human expertise. One model validated in over 16,000 patients demonstrated an AUC of 0.947, with 94.6% sensitivity and 89.0% specificity for detecting cervical lesions.  

In a cohort exceeding 700,000 women, an AI system achieved 94.7% concordance with cytologists while increasing sensitivity by 5.8%. These findings suggest genuine potential to reduce false negatives, the principal limitation of cytology-based screening. Technological approaches are evolving toward multimodal models integrating cytological images, HPV status, and molecular biomarkers. Semi-supervised and self-supervised learning techniques allow the use of large partially annotated image datasets. Additionally, the emergence of explainable AI (xAI) models aims to improve decision transparency, a prerequisite for clinical acceptance.  

Commercial solutions such as the Genius™ Digital Diagnostics System, already authorized by the FDA, illustrate the shift from experimental development to routine implementation. AI is no longer confined to research; it is becoming an operational tool for case triage and prioritization.  

Screening enters the intelligent era

Cervical cancer remains preventable through organized screening. However, the real-world effectiveness of such programs depends on the quality and reproducibility of cytological interpretation, still subject to human variability and significant analytical workload.  

The current challenge extends beyond lesion detection to include standardization, reliability, and optimization of diagnostic workflows. This review aimed to evaluate the maturity and clinical relevance of AI applied to cervical cytology by analyzing its diagnostic performance and integration potential in routine practice.  

Available data indicate that AI significantly improves diagnostic sensitivity, reduces inter-observer variability, and contributes to standardization of morphological assessment, with robust performance in large cohorts. These findings support its role as a triage-support tool and as a means of strengthening screening safety.  

Although further multicenter validation in more diverse populations is required, the evidence highlights the potential of AI-assisted cytology to consolidate the performance of prevention programs. Ultimately, this approach may enable finer risk stratification and greater efficiency in diagnostic pathways, thereby reinforcing the long-term impact of cervical cancer screening.



                        Read next: Stem cell transplantation: ai to predict severe complications  



About the author – Ana Espino
PhD in Immunology, specialized in Virology  
As a scientific writer, Ana is passionate about bridging the gap between research and real-world impact. With expertise in immunology, virology, oncology, and clinical studies, she makes complex science clear and accessible. Her mission: to accelerate knowledge sharing and empower evidence-based decisions through impactful communication.