A method for screening false positive of cervical abnormal cells based on morphological parameter constraint

CN122265726APending Publication Date: 2026-06-23NORTHEAST FORESTRY UNIV

View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office: CN · China
Patent Type: Applications(China)
Current Assignee / Owner: NORTHEAST FORESTRY UNIV
Filing Date: 2026-04-02
Publication Date: 2026-06-23

AI Technical Summary

Technical Problem

Existing deep learning models have a high false positive rate in cervical cell smear detection, making it difficult to effectively distinguish between lesion cells and background noise. Furthermore, their generalization ability is poor when applied across centers, leading to an increased workload for pathologists.

Method used

A morphological parameter-constrained cervical abnormal cell screening method is adopted, including overlapping sliding window segmentation, multi-dimensional image quality assessment, soft nonmaximum suppression, morphological mask extraction, and comprehensive discriminant function. Combined with an unsupervised heuristic pseudo-label optimization mechanism, the weights and thresholds are dynamically adjusted.

Benefits of technology

It significantly reduces the false positive rate, improves the recall rate of real lesion cells, reduces the cost of manual annotation, and realizes efficient and high-precision automated assisted screening of full-view digital slides of cervical liquid-based cytology.

✦ Generated by Eureka AI based on patent content.

Smart Images

Figure CN122265726A_ABST

Patent Text Reader

Abstract

The present application aims at the problem that the existing cervical cell auxiliary screening is easily interfered by complex background, resulting in high false positive rate and often missing dense overlapping cells. A false positive screening method for cervical abnormal cells based on morphological parameter constraint is disclosed. The method first acquires a full field digital slice and extracts an original image block; then outputs a prediction box through a target detection network, and uses soft non-maximum suppression to retain closely arranged cell clusters; then performs morphological double mask extraction on the candidate region to separate the cell plasma and nucleus; further calculates morphological features such as nuclear-cytoplasmic ratio and nuclear roundness, and introduces a fault-tolerant and truncation mechanism to ensure numerical stability; finally, a multi-dimensional comprehensive discriminant function is constructed to calculate the abnormal score, and an unsupervised heuristic pseudo-label optimization mechanism is combined to dynamically determine the judgment threshold, so as to eliminate false positive targets caused by impurities, inflammation or tissue overlap. The accuracy and robustness of the full automatic screening of cervical cells are effectively improved.

Need to check novelty before this filing date? Find Prior Art