A Study on the Application of an AI Image Recognition-Based Early Cervical Cancer Screening System

Abstract

Objective: To investigate the feasibility and value of a novel early cervical cancer screening system based on Artificial Intelligence (AI) and super-high-speed cell imaging technology in clinical applications, with a focus on evaluating its sensitivity in identifying positive samples. Methods: This study included 2000 samples from women undergoing cervical cancer screening. A self-developed “super-high-speed cell imaging intelligent analyzer” was used for automated image acquisition and multi-dimensional feature extraction of exfoliated cervical cells. Subsequently, a deep learning-based AI model was utilized for intelligent recognition and classification of the cell images. All results judged as positive by the AI system were confirmed by histopathological examination, which served as the gold standard for calculating the system’s diagnostic sensitivity. Results: Among the 2000 total samples, 242 were confirmed as positive by pathology. The AI screening system successfully identified 231 of these cases, missing 11, resulting in a sensitivity of 95.45% for identifying positive samples. Analysis revealed that the system has stable and efficient recognition capabilities for mid-to-high-grade lesions, while all missed cases were low-grade lesions with a scarce number of abnormal cells and atypical features. Conclus ion: This study confirms that the early cervical cancer screening system based on super-high-speed cell imaging and AI recognition has extremely high sensitivity. Its automation and high-throughput characteristics can effectively compensate for the shortcomings of traditional screening methods, such as strong subjectivity, reliance on manual labor, and insufficient pathologist resources in primary care. Although there is still room for improvement in identifying very early or low-burden lesions, and the system currently serves mainly as an efficient triage tool, it has demonstrated great potential for application in large-scale cervical cancer screening. Future optimization will focus on enhancing the recognition of low-grade lesions and developing a multi-class AI model capable of lesion subtype grading to better assist in precise clinical diagnosis.