Related Experiment Video
Updated: Sep 3, 2025
07:23
Retinal Vascular Reactivity as Assessed by Optical Coherence Tomography Angiography
Published on: March 26, 2020
7.6K
HCTNet: A Hybrid ConvNet-Transformer Network for Retinal Optical Coherence Tomography Image Classification
Zongqing Ma1,2, Qiaoxue Xie1,2, Pinxue Xie3
1Key Laboratory of the Ministry of Education for Optoelectronic Measurement Technology and Instrument, Beijing Information Science and Technology University, Beijing 100192, China.
Biosensors
|July 27, 2022
Summary
A new hybrid deep learning model, HCTNet, accurately classifies retinal optical coherence tomography (OCT) images for disease diagnosis. This Transformer-ConvNet approach outperforms existing methods in computer-assisted diagnosis.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Accurate classification of optical coherence tomography (OCT) images is crucial for computer-assisted diagnosis of retinal diseases.
- Existing methods may struggle to capture both local and global image features effectively.
Purpose of the Study:
- To propose and validate a hybrid ConvNet-Transformer network (HCTNet) for enhanced retinal OCT image classification.
- To demonstrate the efficacy of Transformer-based approaches in this diagnostic task.
Main Methods:
- Developed a hybrid ConvNet-Transformer network (HCTNet) integrating residual dense blocks for low-level feature extraction.
- Employed parallel Transformer and ConvNet branches to capture global and local image contexts, respectively.
- Utilized an adaptive re-weighting mechanism for feature fusion to predict OCT image categories.
Main Results:
- The HCTNet achieved high classification accuracies of 91.56% and 86.18% on two public retinal OCT datasets.
- The proposed HCTNet method outperformed pure Vision Transformer (ViT) and several Convolutional Neural Network (ConvNet)-based classification approaches.
- Demonstrated the synergistic benefits of combining local feature extraction (ConvNet) with long-range dependency modeling (Transformer).
Conclusions:
- The hybrid ConvNet-Transformer network (HCTNet) offers a powerful and accurate solution for retinal OCT image classification.
- Transformer-based methods show significant potential for advancing computer-assisted diagnosis in ophthalmology.
- HCTNet effectively leverages the complementary strengths of ConvNets and Transformers for improved diagnostic performance.
