Diagnosis of Coronavirus Disease 2019 (COVID-19) with Structured Latent Multi-View Representation Learning

Abstract: Recently, the outbreak of Coronavirus Disease 2019 (COVID-19) has spread rapidly across the world. Due to the large number of affected patients and heavy labor for doctors, computer-aided diagnosis with machine learning algorithm is urgently needed, and could largely reduce the efforts of clinicians and accelerate the diagnosis process. Chest computed tomography (CT) has been recognized as an informative tool for diagnosis of the disease. In this study, we propose to conduct the diagnosis of COVID-19 with a series of features extracted from CT images. To fully explore multiple features describing CT images from different views, a unified latent representation is learned which can completely encode information from different aspects of features and is endowed with promising class structure for separability. Specifically, the completeness is guaranteed with a group of backward neural networks (each for one type of features), while by using class labels the representation is enforced to be compact within COVID-19/community-acquired pneumonia (CAP) and also a large margin is guaranteed between different types of pneumonia. In this way, our model can well avoid overfitting compared to the case of directly projecting highdimensional features into classes. Extensive experimental results show that the proposed method outperforms all comparison methods, and rather stable performances are observed when varying the numbers of training data.

Diagnosis of COVID-19 with Structured Latent Multi-View Representation Learning

The paper "Diagnosis of Coronavirus Disease 2019 (COVID-19) with Structured Latent Multi-View Representation Learning," presents an innovative approach for COVID-19 diagnosis using chest computed tomography (CT) images through structured latent multi-view representation learning. This method effectively incorporates multiple feature types extracted from CT images to assist in discerning COVID-19 from community-acquired pneumonia (CAP).

Methodological Overview

The researchers employ a multi-view learning approach to capture the complementary aspects of diverse feature sets, namely radiomic and handcrafted features. Radiomic features include gray and texture features, which are derived from various matrices to represent intensity distributions and textural details in the images. Handcrafted features consist of measures like histogram, number, intensity, surface, and volume features, all aimed at characterizing the infection areas in the lungs. A total of 189-dimensional features are extracted, divided into 7 distinct feature groups.

The proposed diagnosis framework is divided into three steps:

1. Complete and Structured Representation Learning: This involves learning a latent representation that integrates information from different feature views while ensuring separability based on class labels. Completeness of the latent representation is ensured by reconstructing different feature types through backward neural networks.
2. Latent-Representation Regressor: To maintain consistency between training and testing stages, a regression model is trained to project original feature sets into the learned latent space.
3. Latent-Representation-Based Classifier: A classifier is trained on these latent representations to differentiate between COVID-19 and CAP effectively.

Experimental Details

The dataset comprises 2522 CT images, among which 1495 cases are COVID-19 positive and 1027 cases are CAP. The experimental results reveal that the latent representation significantly enhances the diagnostic performance compared to using original features directly. The achieved accuracy, sensitivity, and specificity substantiate the robustness of the approach. Specifically, a diagnosis accuracy of 95.50% was achieved, with sensitivity and specificity of 96.6% and 93.2%, respectively.

Key Findings and Implications

1. Multi-view Feature Utility: The study highlights the importance of integrating multiple feature types for improved diagnostic accuracy. Radiomic features exhibited strong discriminative power compared to individual handcrafted features, yet a combined approach leveraged complementary information across all feature types effectively.
2. Stability and Generalization: The proposed method demonstrated stability across various training dataset sizes, indicating adaptability to different data availability conditions—a crucial consideration in clinical applications.
3. Latent Representation Superiority: By projecting features into a low-dimensional space that maintains class separability, the latent representation avoids overfitting and improves generalization, outperforming direct feature-based methods.

Future Considerations

The framework demonstrates a promising pathway for automated COVID-19 diagnosis and emphasizes the potential of multi-view learning techniques in medical image analysis. Future investigations could include expanding the classification to encompass varying severities of COVID-19 and integrating additional clinical data to further boost diagnostic accuracy. Moreover, adaptations in multi-class classification scenarios could be explored to broaden the applicability of the method in differential diagnosis settings.

Overall, this study contributes significantly to the field of computer-aided diagnosis by effectively leveraging structured latent space representations and multi-view learning to enhance the diagnostic process in the context of COVID-19 and related pulmonary conditions.