The worldwide pandemic caused by the new coronavirus (COVID-19) has encouraged the development of multiple computer-aided diagnosis systems to automate daily clinical tasks, such as abnormality detection and classification. Among these tasks, the segmentation of COVID lesions is of high interest to the scientific community, enabling further lesion characterization. Automating the segmentation process can be a useful strategy to provide a fast and accurate second opinion to the physicians, and thus increase the reliability of the diagnosis and disease stratification.
The current work explores a CNN-based approach to segment multiple COVID lesions. It includes the implementation of a U-Net structure with a ResNet34 encoder able to deal with the highly imbalanced nature of the problem, as well as the great variability of the COVID lesions, namely in terms of size, shape, and quantity. This 2D approach yields a Dice score of 64.1%, when evaluated on the publicly available COVID-19-20 Lung CT Lesion Segmentation GrandChallenge data set.
This code was developed using a Keras framework with a Tensorflow backend. The file with all the requirements is included in the repository (requirements_env.yml).
1_preprocessing.py - Receives original dataset images and applies preprocessing steps according to what is described in the publication.
2_train_evaluate_model.py - Optimizes and evaluates the U-Net model using the preprocessed images.
3_visualize_results.py - Creates an overlayed plot of the ground truth and predicted segmentation images, marking the True Positive pixels in green, False Negative pixels in red and False Positive pixels in yellow.
4_get_metrics.py - Based on the ground truth and predicted segmentation images, calculates the accuracy, accuracy per class, recall, precision, dice, and jaccard metrics.
If you use this code, please cite the following publication: J. Rocha, S. Pereira, A. Campilho and A. M. Mendonça, "Segmentation of COVID-19 Lesions in CT Images," 2021 IEEE EMBS International Conference on Biomedical and Health Informatics (BHI), 2021, pp. 1-4, doi: 10.1109/BHI50953.2021.9508520. https://ieeexplore.ieee.org/document/9508520