Related Experiment Video
Updated: Jul 16, 2026
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
Systematic Evaluation of Vision Transformers for Automated Cervical Cancer Classification: Optimization, Statistical
Nisreen Albzour1, Sarah S Lam1
1School of Systems Science and Industrial Engineering, Binghamton University, Binghamton, NY 13902, USA.
Cancers
|July 15, 2026
Summary
Vision Transformers (ViT) show promise for automated cervical cancer screening, achieving ~95% accuracy. This AI approach offers interpretable insights by focusing on clinically relevant cell features, improving upon traditional methods.
Area of Science:
- Medical AI
- Computer Vision
- Oncology
Background:
- Manual Pap smear analysis for cervical cancer screening faces challenges including inter-observer variability and limited expert availability.
- Existing automated methods using convolutional neural networks (CNNs) struggle with long-range spatial dependencies and clinical interpretability.
Purpose of the Study:
- To systematically optimize Vision Transformer (ViT) architectures for enhanced automated cervical cancer screening.
- To improve the interpretability of AI models in cervical cancer diagnosis.
Main Methods:
- Utilized the Herlev dataset (917 images) comprising normal and abnormal cervical cells.
- Optimized a lightweight ViT architecture (ViT-Tiny) by evaluating augmentation strategies, class weighting, and hyperparameters.
- Employed Gradient-weighted Class Activation Mapping (Grad-CAM) for model interpretability analysis.
Main Results:
- Achieved a cross-validation accuracy of approximately 95% with optimal configuration.
- Identified random horizontal flipping and specific class weighting (0.7 × 1.3) as most effective.
- Grad-CAM analysis demonstrated model attention aligned with clinically relevant cytopathological features like nuclei and chromatin texture.
Conclusions:
- Vision Transformers offer accurate and interpretable decision support for cervical cancer screening.
- The attention-based transparency of ViTs is relevant for medical AI applications.
- Further validation on larger, multi-center datasets is required prior to clinical deployment.
