Dual-Sampling Attention Network for Diagnosis of COVID-19 From Community Acquired Pneumonia

Ouyang, Xin; Huo, Jiayu; Xia, Liming; Shan, Fei; Liu, Jun; Mo, Zhanhao; Yan, Fuhua; Ding, Zhongxiang; Yang, Qi; Song, Bin; Shi, Feng; Yuan, Huan; Wei, Ying; Cao, Xiaohuan; Gao, Yaozong; Wu, Dijia; Wang, Qian; Shen, Dinggang

doi:10.1109/tmi.2020.2995508

Cited by 331 publications

(262 citation statements)

References 51 publications

(65 reference statements)

Supporting

Mentioning

260

Contrasting

Unclassified

Order By: Relevance

“…With the rapid growth rate of COVID-19 suspection all over the world, designing effective automated tools for COVID-19 diagnosis from CT imaging is highly demanded to improve the clinical diagnosis efficiency and release the tedious workload of clinicians and radiologists. However, accurate diagnosis of COVID-19 from CT images is a non-trival problem, mainly due to the highly similar patterns of COVID-19 and other pneumonia types, as well as the large appearance variance of COVID-19 lesions of patients in different severity level [42]. Recently, a variety of data-driven models have been proposed to solve this problem [4], [19], [43], [44], leading to considerable progress in the field of automated COVID-19 diagnosis in the past few months.…”

Section: Discussionmentioning

confidence: 99%

Contrastive Cross-Site Learning With Redesigned Net for COVID-19 CT Classification

Wang

Liu

Dou

2020

IEEE J. Biomed. Health Inform.

154

135

View full text Add to dashboard Cite

The pandemic of coronavirus disease 2019 (COVID-19) has lead to a global public health crisis spreading hundreds of countries. With the continuous growth of new infections, developing automated tools for COVID-19 identification with CT image is highly desired to assist the clinical diagnosis and reduce the tedious workload of image interpretation. To enlarge the datasets for developing machine learning methods, it is essentially helpful to aggregate the cases from different medical systems for learning robust and generalizable models. This paper proposes a novel joint learning framework to perform accurate COVID-19 identification by effectively learning with heterogeneous datasets with distribution discrepancy. We build a powerful backbone by redesigning the recently proposed COVID-Net in aspects of network architecture and learning strategy to improve the prediction accuracy and learning efficiency. On top of our improved backbone, we further explicitly tackle the cross-site domain shift by conducting separate feature normalization in latent space. Moreover, we propose to use a contrastive training objective to enhance the domain invariance of semantic embeddings for boosting the classification performance on each dataset. We develop and evaluate our method with two public large-scale COVID-19 diagnosis datasets made up of CT images. Extensive experiments show that our approach consistently improves the performanceson both datasets, outperforming the original COVID-Net trained on each dataset by 12.16% and 14.23% in AUC respectively, also exceeding existing state-of-the-art multi-site learning methods.

show abstract

Section: Discussionmentioning

confidence: 99%

Contrastive Cross-Site Learning With Redesigned Net for COVID-19 CT Classification

Wang

Liu

Dou

2020

IEEE J. Biomed. Health Inform.

154

135

View full text Add to dashboard Cite

show abstract

“…Deep learning has been employed for the diagnosis of COVID-19 in chest CT Song et al (2020); Gozes et al (2020a,b) and community acquired pneumonia (CAP) Kermany et al (2018). For example, Ouyang et al (2020) proposed a 3D convolutional neural network (CNN) with online attention refinement to diagnose COVID-19 from CAP and introduced a sampling strategy to mitigate the imbalanced distribution of infected regions between COVID-19 and CAP. Song et al (2020) proposed DeepPneumonia for localization and detection of COVID-19 pneumonia; attention was also applied to detect key regions with impressive results on a large cohort.…”

Section: Introductionmentioning

confidence: 99%

Dual Attention Multiple Instance Learning with Unsupervised Complementary Loss for COVID-19 Screening

Chikontwe

Luna

Kang

et al. 2020

Preprint

View full text Add to dashboard Cite

Chest computed tomography (CT) based analysis and diagnosis of the Coronavirus Disease 2019 (COVID-19) plays a key role in combating the outbreak of the pandemic that has rapidly spread worldwide. To date, the disease has infected more than 18 million people with over 690k deaths reported. Reverse transcription polymerase chain reaction (RT-PCR) is the current gold standard for clinical diagnosis but may produce false positives; thus, chest CT based diagnosis is considered more viable. However, accurate screening is challenging due to difficulty in annotation efforts of infected areas, curation of large datasets, and the slight discrepancies between COVID-19 and other viral pneumonia. In this study, we propose an attention-based end-to-end weakly supervised framework for the rapid diagnosis of COVID-19 and bacterial pneumonia based on multiple instance learning (MIL). We further incorporate unsupervised contrastive learning for improved accuracy with attention applied both in spatial and latent contexts, herein we propose Dual Attention Contrastive based MIL (DA-CMIL). DA-CMIL takes as input a several patient CT slices (considered as a bag of instances) and outputs a single label. Attention based pooling is applied to implicitly select key slices in latent space, and spatial attention learns slice spatial context for interpretable diagnosis. A contrastive loss is applied at the instance level to encode similarity in features from the same patient against pooled patient features. Empirical results show our algorithm achieves an overall accuracy of 98.6% and an AUC of 98.4%. Moreover, ablation studies show the benefit of contrastive learning with MIL.

show abstract

“…To mitigate the burden on radiologists, while providing the highest quality care for patients, there has been tremendous effort to develop novel image processing approaches using machine learning algorithms 24 , particularly for COVID-19 diagnosis and prognosis 25 . These artificial intelligence (AI) models exploit and build upon medical imaging modalities such as chest CT scans [26][27][28][29][30][31][32] , chest radiographs [33][34][35][36][37][38][39][40] , and lung ultrasound 41 However, for any of these AI models to be useful in assisting clinicians in the care of COVID-19 patients, they require a robust and reliable AI deployment system 42 . Deployment is often a difficult step because clinical radiology infrastructure is not designed for easily embedding third-party systems, and doing so while maintaining context sensitivity and seamlessly embedding such systems into the radiologist workflow requires knowledge of hospital information system integration standards and often product-specific knowledge.…”

Section: Introductionmentioning

confidence: 99%

Clinical deployment and validation of a radiology artificial intelligence system for COVID-19

Hashemian

Manchanda

et al. 2020

Preprint

View full text Add to dashboard Cite

The global COVID-19 pandemic has disrupted patient care delivery in healthcare systems world-wide. For healthcare providers to better allocate their resources and improve the care for patients with severe disease, it is valuable to be able to identify those patients with COVID-19 who are at higher risk for clinical complications. This may help to optimize clinical workflow and more efficiently allocate scarce medical resources. To this end, medical imaging shows great potential and artificial intelligence (AI) algorithms have been developed to assist in diagnosing and risk stratifying COVID-19 patients. However, despite the rapid development of numerous AI models, these models cannot be clinically useful unless they can be deployed in real-world environments in real-time on clinical data. Here, we propose an end-to-end AI hospital-deployment architecture for COVID-19 medical imaging algorithms in hospitals. We have successfully implemented this system at our institution and it has been used in prospective clinical validation of a deep learning algorithm potentially useful for triaging of patients with COVID-19. We demonstrate that many orchestration processes are required before AI inference can be performed on a radiology studies in real-time with the AI model being just one of the components that make up the AI deployment system. We also highlight that failure of any one of these processes can adversely affect the model's performance.

show abstract

Dual-Sampling Attention Network for Diagnosis of COVID-19 From Community Acquired Pneumonia

Cited by 331 publications

References 51 publications

Contrastive Cross-Site Learning With Redesigned Net for COVID-19 CT Classification

Contrastive Cross-Site Learning With Redesigned Net for COVID-19 CT Classification

Dual Attention Multiple Instance Learning with Unsupervised Complementary Loss for COVID-19 Screening

Clinical deployment and validation of a radiology artificial intelligence system for COVID-19

Contact Info

Product

Resources

About