JCS: An Explainable COVID-19 Diagnosis System by Joint Classification and Segmentation (2004.07054v3)

Abstract: Recently, the coronavirus disease 2019 (COVID-19) has caused a pandemic disease in over 200 countries, influencing billions of humans. To control the infection, identifying and separating the infected people is the most crucial step. The main diagnostic tool is the Reverse Transcription Polymerase Chain Reaction (RT-PCR) test. Still, the sensitivity of the RT-PCR test is not high enough to effectively prevent the pandemic. The chest CT scan test provides a valuable complementary tool to the RT-PCR test, and it can identify the patients in the early-stage with high sensitivity. However, the chest CT scan test is usually time-consuming, requiring about 21.5 minutes per case. This paper develops a novel Joint Classification and Segmentation (JCS) system to perform real-time and explainable COVID-19 chest CT diagnosis. To train our JCS system, we construct a large scale COVID-19 Classification and Segmentation (COVID-CS) dataset, with 144,167 chest CT images of 400 COVID-19 patients and 350 uninfected cases. 3,855 chest CT images of 200 patients are annotated with fine-grained pixel-level labels of opacifications, which are increased attenuation of the lung parenchyma. We also have annotated lesion counts, opacification areas, and locations and thus benefit various diagnosis aspects. Extensive experiments demonstrate that the proposed JCS diagnosis system is very efficient for COVID-19 classification and segmentation. It obtains an average sensitivity of 95.0% and a specificity of 93.0% on the classification test set, and 78.5% Dice score on the segmentation test set of our COVID-CS dataset. The COVID-CS dataset and code are available at https://github.com/yuhuan-wu/JCS.

• The paper presents a novel framework that combines classification and segmentation to analyze chest CT scans for COVID-19 diagnosis.
• It leverages the extensive COVID-CS dataset and activation mapping to achieve 95% sensitivity and 93% specificity in detection.
• The integration of patient-level and pixel-level annotations mitigates data bias and enhances lesion quantification compared to traditional models.

Overview of "JCS: An Explainable COVID-19 Diagnosis System by Joint Classification and Segmentation"

The paper "JCS: An Explainable COVID-19 Diagnosis System by Joint Classification and Segmentation" presents an innovative system aimed at enhancing COVID-19 diagnostic capabilities using chest CT scans. This system, termed JCS, integrates joint classification and segmentation methods to provide real-time and explainable diagnosis of COVID-19, addressing the limitations of existing diagnostic modalities such as RT-PCR tests and manual CT evaluations performed by radiologists.

Key Contributions

1. Dataset Construction: A notable contribution of this research is the creation of the COVID-19 Classification and Segmentation (COVID-CS) dataset. This extensive dataset includes 144,167 chest CT images sourced from 750 cases, with 400 confirmed COVID-19 patients. It includes 3,855 images annotated with pixel-level labels, providing a substantial resource for training deep learning models.
2. Joint Classification and Segmentation System: The JCS system leverages convolutional neural networks (CNNs) to classify CT images as either COVID-19 positive or negative. It provides evidence of classification through activation mapping techniques, enhancing the transparency and interpretability of the model's predictions. A segmentation model is further employed to delineate and quantify the opacification regions indicative of COVID-19 infection, allowing for detailed assessment of disease severity.
3. Methodological Innovations: The system addresses potential data biases by using image mixing techniques to mitigate overfitting and ensure robust detection of lesions. Moreover, by utilizing a combination of both patient-level and pixel-level annotations during training, the system achieves significant improvements in prediction accuracy.
4. Performance Statistics: The JCS system demonstrates high diagnostic performance with an average sensitivity of 95.0% and specificity of 93.0% on the classification dataset. For segmentation, it achieves a Dice score of 78.5%, outperforming existing methods like U-Net, DSS, and PoolNet in accurately delineating infected lung regions.

Implications and Future Directions

The implications of this work span both practical applications in clinical diagnostic workflows and advancements in AI methodologies for medical imaging. From a practical standpoint, the JCS system significantly reduces the time required to diagnose COVID-19 via CT scans, thereby potentially alleviating the burden on healthcare systems during pandemics.

Theoretically, the integration of joint classification and segmentation within a single framework opens avenues for more sophisticated diagnostic tools that can be applied to other medical imaging tasks beyond COVID-19. Additionally, the development of large-scale annotated datasets, such as COVID-CS, sets a precedent for future data collection efforts in emerging medical challenges.

Future research could aim at integrating newer AI paradigms, such as transformers or neural architecture search (NAS), into the existing framework to further enhance performance. Moreover, exploring the insights derived from explainable AI mechanisms in refining and validating medical imaging models remains an intriguing direction. The continual expansion and validation of datasets will also be essential in adapting the framework to detect future variants of COVID-19 or other respiratory diseases.

In summary, this paper presents a substantial advancement in using AI for COVID-19 diagnosis, combining rigorous methodological frameworks with practical benefits, and lays the groundwork for future developments in intelligent medical imaging solutions.