Related Experiment Video
Updated: Sep 20, 2025
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
2.0K
Bladder lesion detection using EfficientNet and hybrid attention transformer through attention transformation
Poonam Sharma1, Bhisham Sharma2, Dhirendra Prasad Yadav3
1Chitkara University Institute of Engineering and Technology, Chitkara University, Rajpura, Punjab, 140401, India.
Scientific Reports
|May 23, 2025
Summary
This study introduces a hybrid Convolutional Neural Network (CNN) and Vision Transformer (ViT) model for accurate bladder cancer diagnosis from endoscopic images. The novel approach significantly improves diagnostic performance over existing methods.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Bladder cancer diagnosis is complex due to tumor heterogeneity and cell morphology, making manual methods time-consuming.
- Current machine learning and deep learning methods face limitations in real-time application due to manual feature engineering or extensive data requirements.
Purpose of the Study:
- To develop a hybrid Convolutional Neural Network (CNN) and Vision Transformer (ViT) model for efficient and accurate bladder cancer diagnosis.
- To overcome the limitations of traditional machine learning and deep learning approaches in real-time bladder lesion detection.
Main Methods:
- A hybrid model combining InceptionV3 blocks for spatial feature extraction and a Vision Transformer (ViT) encoder with hybrid attention modules for global feature correlation was developed.
- The model was trained and evaluated on a dataset of 17,540 endoscopic images using a 5-fold cross-validation strategy.
Main Results:
- The proposed hybrid CNN-ViT model achieved high performance metrics: 97.73% average accuracy, 97.21% precision, and 96.86% F1-score.
- Comparative analysis demonstrated superior performance of the hybrid model over standalone CNN and ViT-based methods under identical experimental conditions.
Conclusions:
- The developed hybrid CNN-ViT model offers a promising, reliable, and efficient solution for automated bladder cancer diagnosis from endoscopic images.
- This approach addresses the challenges of manual diagnosis and limitations of existing AI methods, paving the way for improved clinical applications.
