A COVID-19 medical image classification algorithm based on Transformer.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Deep Learning

Background:

• Coronavirus 2019 (COVID-19) is a rapidly spreading acute respiratory illness.
• Accurate and efficient diagnostic tools are crucial for managing the pandemic.

Purpose of the Study:

• To propose a novel deep learning network, RMT-Net, for detecting and classifying COVID-19 from medical images.
• To evaluate the performance of RMT-Net against existing models.

Main Methods:

• Developed RMT-Net by merging ResNet-50 with Transformer architecture for enhanced feature extraction.
• Utilized convolutional neural networks and depth-wise convolution for local features and computational efficiency.
• Employed global self-attention and residual blocks for capturing both long-distance and detailed features.

Main Results:

• RMT-Net achieved high accuracy: 97.65% on X-ray datasets and 99.12% on CT datasets.
• The model demonstrated superior performance compared to ResNet-50, VGGNet-16, i-CapsNet, and MGMADS-3.
• RMT-Net is computationally efficient with a model size of 38.5 M and fast detection speeds (5.46 ms for X-ray, 4.12 ms for CT).

Conclusions:

• RMT-Net effectively detects and classifies COVID-19 with high accuracy and efficiency.
• The proposed model offers a promising solution for rapid and reliable COVID-19 diagnosis using medical imaging.