Dual-Sampling Attention Network for Diagnosis of COVID-19 from Community Acquired Pneumonia (2005.02690v2)

Abstract: The coronavirus disease (COVID-19) is rapidly spreading all over the world, and has infected more than 1,436,000 people in more than 200 countries and territories as of April 9, 2020. Detecting COVID-19 at early stage is essential to deliver proper healthcare to the patients and also to protect the uninfected population. To this end, we develop a dual-sampling attention network to automatically diagnose COVID- 19 from the community acquired pneumonia (CAP) in chest computed tomography (CT). In particular, we propose a novel online attention module with a 3D convolutional network (CNN) to focus on the infection regions in lungs when making decisions of diagnoses. Note that there exists imbalanced distribution of the sizes of the infection regions between COVID-19 and CAP, partially due to fast progress of COVID-19 after symptom onset. Therefore, we develop a dual-sampling strategy to mitigate the imbalanced learning. Our method is evaluated (to our best knowledge) upon the largest multi-center CT data for COVID-19 from 8 hospitals. In the training-validation stage, we collect 2186 CT scans from 1588 patients for a 5-fold cross-validation. In the testing stage, we employ another independent large-scale testing dataset including 2796 CT scans from 2057 patients. Results show that our algorithm can identify the COVID-19 images with the area under the receiver operating characteristic curve (AUC) value of 0.944, accuracy of 87.5%, sensitivity of 86.9%, specificity of 90.1%, and F1-score of 82.0%. With this performance, the proposed algorithm could potentially aid radiologists with COVID-19 diagnosis from CAP, especially in the early stage of the COVID-19 outbreak.

• The paper presents a novel dual-sampling strategy with an online attention module that enhances CT-based COVID‑19 diagnosis.
• The paper achieves robust performance with an AUC of 0.944, 87.5% accuracy, and high sensitivity and specificity in differentiating COVID‑19 from CAP.
• The paper improves model interpretability and addresses data imbalance, providing a practical tool to aid radiologists in rapid diagnosis.

Overview of Dual-Sampling Attention Network for Identifying COVID-19 via CT Imaging

This paper presents a paper focused on the automated diagnosis of COVID-19 from Community Acquired Pneumonia (CAP) using chest computed tomography (CT) scans. The proposed methodology introduces a Dual-Sampling Attention Network that leverages a novel combination of a dual-sampling strategy and an online attention mechanism integrated with a 3D convolutional neural network.

Methodology

The framework comprises two major components: a dual-sampling strategy and an online attention module.

1. Online Attention Module: This component utilizes sophisticated attention mechanisms integrated with CAM (Class Activation Mapping) extended for 3D data. The attention maps are dynamically refined using segmentation masks of infection regions to focus the network's attention on critical areas within the lungs during the training phase. This mechanism addresses the model interpretability challenge by providing insights into the areas influencing the predictions.
2. Dual-Sampling Strategy: It incorporates a size-balanced sampling technique to counteract the data imbalance inherent in COVID-19 and CAP infection region sizes. This involves boosting sampling probabilities for cases with rare features, such as COVID-19 instances with small infection areas or CAP instances with large infection areas.

The paper implements the ensemble learning approach to aggregate predictions from models trained under different sampling techniques, thereby combining the benefits of both uniform and size-balanced sampling to deliver enhanced performance.

Experimental Results

The research utilizes comprehensive datasets, having 4982 CT scans across consecutive stages of training-validation and testing from different hospitals. During evaluations:

• The model demonstrated a robust ability to differentiate between COVID-19 and CAP, achieving an AUC of 0.944, an accuracy of 87.5%, sensitivity of 86.9%, specificity of 90.1%, and an F1-score of 82.0%.
• The model's reliability was further corroborated by utilizing an independent testing dataset, which reinforced the generalization capabilities of the method.

The paper meticulously compares the proposed model's performance to baseline configurations, such as the traditional 3D ResNet34 networks without the dual-sampling strategy or attention module, establishing its effectiveness in improving the diagnostic accuracy, particularly for cases with distinct infection region characteristics.

Implications and Future Directions

The implications of this work are both practical and strategic in the domain of medical imaging and diagnostics:

• Practical Impact: The methodology can significantly aid radiologists by providing automated and reliable differentiation of COVID-19 from CAP, which is particularly beneficial during pandemic outbreaks when rapid, scalable diagnostic tools are essential.
• Theoretical Insights: This research emphasizes the importance of balance in training data distribution and attention mechanisms to enhance model interpretability and diagnostic performance in medical image analysis. It opens avenues for further enhancements in deep learning frameworks tailored for medical diagnostics.

Future work suggested by the paper includes refining attention mechanisms for better visibility into diagnostic features, expanding datasets for even more diverse generalization, and exploring longitudinal studies to track disease progression. Additionally, integrating the model output with clinical and laboratory data is recommended to improve diagnostic precision.

Overall, this paper contributes to the field by demonstrating a methodologically rigorous approach to solving the challenging problem of COVID-19 diagnosis through CT imaging, offering a potentially valuable tool to supplement traditional diagnostic methods.