Related Experiment Video
Updated: Jan 17, 2026
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
2.3K
HMCFormer (hierarchical multi-scale convolutional transformer): a hybrid CNN+Transformer network for intelligent VIA
Bo Feng1,2, Chao Xu1,2, Zhengping Li1,2
1School of Integrated Ciruits, Anhui University, HeFei, Anhui, China.
Peerj. Computer Science
|September 24, 2025
Summary
A new AI model, the Hierarchical Multi-Scale Convolutional Transformer (HMCFormer), improves cervical cancer screening accuracy using visual inspection with acetic acid (VIA). This technology offers a cheaper, faster method for early detection, especially benefiting low-income populations.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Cervical cancer is a leading cause of cancer death among women globally, particularly in developing nations.
- Visual inspection with acetic acid (VIA) is a cost-effective screening method, but AI-assisted diagnosis faces challenges with accuracy and equipment costs.
- Existing AI-based VIA screening methods often lack sufficient accuracy or require expensive supplementary equipment.
Purpose of the Study:
- To develop an advanced AI model for accurate and accessible cervical cancer screening using VIA.
- To address the limitations of current AI-assisted VIA screening, focusing on improving accuracy and reducing equipment dependency.
- To enhance early detection of cervical cancer through an intelligent and cost-effective screening solution.
Main Methods:
- Proposed the Hierarchical Multi-Scale Convolutional Transformer (HMCFormer) network, integrating CNNs for hierarchical feature extraction and Transformers for global dependency modeling.
- Developed a dual-color space-based image enhancement algorithm for the CNN branch and a hierarchical multi-scale pixel excitation module for adaptive feature extraction.
- Introduced adaptive preprocessing and superiority-inferiority fusion concepts for improved collaboration between the Transformer and CNN branches, utilizing the Swin Transformer architecture.
Main Results:
- The HMCFormer achieved a high screening accuracy of 97.4% on the newly created PCC5000 dataset.
- The model demonstrated a grading accuracy of 94.8% on the same dataset.
- The proposed feature fusion module significantly enhanced the synergistic performance of the network's branches.
Conclusions:
- The HMCFormer network represents a significant advancement in AI-assisted VIA screening for cervical cancer.
- The developed model offers a promising, accurate, and potentially low-cost solution for early cervical cancer detection, especially in resource-limited settings.
- This AI approach can help increase participation in regular cervical cancer screening among vulnerable populations.
