Classification of SARS COVID-19 and Other Lung Infections from Chest X-Ray Scan Images with DenseNet-121
This example shows how to train a deep neural network to classify SARS COVID-19 and other lung infections using chest X-ray (CXR) images. In this example, a transfer learning (TL) based approach is proposed and presented to fine-tune the DenseNet-121 neural network [1] for the CXR image classification task. This example performs classification on the COVID-19 Radiography data set [2]-[3] using a 2-D DenseNet-121 architecture.
- MATLAB® R2021a or later.
- Image Processing Toolbox™.
- Deep Learning Toolbox™.
- Deep Learning Toolbox Converter for ONNX Model Format™.
- Parallel Computing Toolbox™ (if using GPU for training).
In comparison to the classification of photographic images, the classification of medical images presents several challenges, such as:
- Unreliable performance due to low inter-class variation between images.
- Potential overfitting due to imbalanced data sets.
Hence, prior to performing TL based classification, it is important to:
- Select the appropriate deep learning architecture.
- Perform the proper data augmentation techniques.
The 2-D DenseNet-121 [1] is a convolutional network model in which each layer is connected to every other layer in a feed-forward fashion. DenseNets have a number of appealing advantages, including the elimination of the vanishing-gradient problem, improved feature propagation, feature reuse, and a significant reduction in parameter counts. It is also fast, efficient, simple and a popular network in the medical imaging domain [2]-[3]. This example starts with the ONNX™ DenseNet-121 model (ONNX™ version 1.4, Opset version 9) and applies transfer learning to retrain the network using CXR images.
Download or clone this repository to your machine and open the .mlx file in MATLAB®.
| Model | Overall Accuracy | Size (MB) | Classes |
|---|---|---|---|
| DenseNet-121 | >90% | 27 | COVID, Lung_Opacity, Normal, Viral Pneumonia |
Four sample test images with their predicted labels and the prediction scores.
Confusion matrix for the true labels targets and predicted labels outputs.
[1] G. Huang, Z. Liu, L. Van Der Maaten, and K. Q. Weinberger, "Densely Connected Convolutional Networks," In Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 2261-2269, 2017, doi: 10.1109/CVPR.2017.243.
[2] M. E. H. Chowdhury, T. Rahman, A. Khandakar, R. Mazhar, M. A. Kadir, Z. B. Mahbub, K. R. Islam, M. S. Khan, A. Iqbal, N. Al-Emadi, M. B. I. Reaz, and M. T. Islam, "Can AI help in screening Viral and COVID-19 pneumonia?," IEEE Access, Vol. 8, pp. 132665 - 132676, 2020, doi: 10.1109/ACCESS.2020.3010287.
[3] T. Rahman, A. Khandakar, Y. Qiblawey, A. Tahir, S. Kiranyaz, S. B. A. Kashem, M. T. Islam, S. A. Maadeed, S. M. Zughaier, M. S. Khan, and M. E. Chowdhury, "Exploring the Effect of Image Enhancement Techniques on COVID-19 Detection using Chest X-ray Images," Computers in Biology and Medicine, Vol. 132, pp. 104319, 2021, doi: 10.1016/j.compbiomed.2021.104319.
Copyright 2021 The MathWorks, Inc.