A two-stage cervical pathology cell detection model based on YOLOv7x and K-means
Siyu Chen, Jingren Wang · Journal of Artificial Intelligence Practice · 2025
As a malignant tumour that seriously threatens women's health, early diagnosis of cervical cancer is crucial to improve the cure rate. Cervical pathological cell detection is a key link in the early diagnosis of cervical cancer, but the traditional detection methods have the problems of low efficiency and accuracy relying on manual work. In this study, we combined the YOLOv7x target detection model with K-means clustering algorithm to construct a two-stage model YOLOv7x-CH for cervical pathological cell detection. To address the problems in the detection of cervical pathological cells of High-grade Squamous Intraepithelial Lesion (HSIL) category, we firstly constructed the To address the problems in the detection of HSIL, we first constructed the YOLOv7x baseline model, extracted the nucleoplasmic ratio features based on fine segmentation and grey scale symbiosis matrix texture features from the HSIL samples, and then constructed the two-stage model YOLOv7x-CH through K-means clustering. The experimental results show that compared with the baseline model, the two-stage model YOLOv7x-CH can improve the accuracy of cervical pathological cell detection from 0.56 to 0.813, and the average accuracy AP from 0.556 to 0.591, which effectively improves the accuracy of cervical pathological cell detection, and provides a more reliable technical support for the early diagnosis of cervical cancer.