Related Experiment Video
Updated: Sep 24, 2025
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
2.0K
Towards more efficient ophthalmic disease classification and lesion location via convolution transformer
Huajie Wen1, Jian Zhao2, Shaohua Xiang2
1College of Big Data and Internet, Shenzhen Technology University, Shenzhen 518118, China; College of Applied Science, Shenzhen University, Shenzhen 518060, China.
Computer Methods and Programs in Biomedicine
|May 7, 2022
Summary
This study introduces a new deep learning method combining convolution and self-attention for better analysis of retina optical coherence tomography (OCT) images. The LLCT model improves disease classification and lesion localization, aiding ophthalmologists.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Retina optical coherence tomography (OCT) images present challenges like speckle noise and subtle features, hindering accurate diagnosis with conventional methods.
- Existing deep learning models struggle with the unique characteristics of OCT images, limiting classification accuracy and diagnostic utility.
Purpose of the Study:
- To develop a novel deep learning method, the lesion-localization convolution transformer (LLCT), for enhanced classification and lesion localization in retina OCT images.
- To address the limitations of traditional deep learning architectures in analyzing noisy and complex OCT image data.
Main Methods:
- A hybrid architecture integrating convolutional neural networks (CNNs) for feature extraction with self-attention mechanisms for global context analysis.
- Customized feature maps from CNNs serve as input sequences for the self-attention network, leveraging both local and global image information.
- A unique lesion localization mechanism utilizes backpropagation gradients combined with forward-propagated global features.
Main Results:
- The LLCT method demonstrated significant improvements: 7.6% in overall accuracy, 10.9% in sensitivity, and 9.2% in specificity compared to existing approaches.
- Effective localization of lesions within OCT images was achieved without requiring pre-labeled lesion location data.
- The model reduces computational complexity in AI-assisted ophthalmic disease analysis.
Conclusions:
- The proposed LLCT method substantially enhances the performance of AI-assisted ophthalmic disease analysis using OCT images.
- This approach offers a significant advancement in classifying and locating ophthalmic diseases, providing valuable assistance to ophthalmologists.
