Adaptive Feature Selection Guided Deep Forest for COVID-19 Classification With Chest CT.

Chest computed tomography (CT) becomes an effective tool to assist the diagnosis of coronavirus disease-19 (COVID-19). Due to the outbreak of COVID-19 worldwide, using the computed-aided diagnosis technique for COVID-19 classification based on CT images could largely alleviate the burden of clinicians. In this paper, we propose an Adaptive Feature Selection guided Deep Forest (AFS-DF) for COVID-19 classification based on chest CT images. Specifically, we first extract location-specific features from CT images. Then, in order to capture the high-level representation of these features with the relatively small-scale data, we leverage a deep forest model to learn high-level representation of the features. Moreover, we propose a feature selection method based on the trained deep forest model to reduce the redundancy of features, where the feature selection could be adaptively incorporated with the COVID-19 classification model. We evaluated our proposed AFS-DF on COVID-19 dataset with 1495 patients of COVID-19 and 1027 patients of community acquired pneumonia (CAP). The accuracy (ACC), sensitivity (SEN), specificity (SPE), AUC, precision and F1-score achieved by our method are 91.79%, 93.05%, 89.95%, 96.35%, 93.10% and 93.07%, respectively. Experimental results on the COVID-19 dataset suggest that the proposed AFS-DF achieves superior performance in COVID-19 vs. CAP classification, compared with 4 widely used machine learning methods.

CT Image Analysis and Clinical Diagnosis of New Coronary Pneumonia Based on Improved Convolutional Neural Network.

In this paper, based on the improved convolutional neural network, in-depth analysis of the CT image of the new coronary pneumonia, using the U-Net series of deep neural networks to semantically segment the CT image of the new coronary pneumonia, to obtain the new coronary pneumonia area as the foreground and the remaining areas as the background of the binary image, provides a basis for subsequent image diagnosis. Secondly, the target-detection framework Faster RCNN extracts features from the CT image of the new coronary pneumonia tumor, obtains a higher-level abstract representation of the data, determines the lesion location of the new coronary pneumonia tumor, and gives its bounding box in the image. By generating an adversarial network to diagnose the lesion area of the CT image of the new coronary pneumonia tumor, obtaining a complete image of the new coronary pneumonia, achieving the effect of the CT image diagnosis of the new coronary pneumonia tumor, and three-dimensionally reconstructing the complete new coronary pneumonia model, filling the current the gap in this aspect, provide a basis to produce new coronary pneumonia prosthesis and improve the accuracy of diagnosis.

Computed Tomography Image Processing Analysis in COVID-19 Patient Follow-Up Assessment.

The rapid worldwide spread of the COVID-19 pandemic has infected patients around the world in a short space of time. Chest computed tomography (CT) images of patients who are infected with COVID-19 can offer early diagnosis and efficient forecast monitoring at a low cost. The diagnosis of COVID-19 on CT in an automated way can speed up many tasks and the application of medical treatments. This can help complement reverse transcription-polymerase chain reaction (RT-PCR) diagnosis. The aim of this work is to develop a system that automatically identifies ground-glass opacity (GGO) and pulmonary infiltrates (PIs) on CT images from patients with COVID-19. The purpose is to assess the disease progression during the patient's follow-up assessment and evaluation. We propose an efficient methodology that incorporates oversegmentation mean shift followed by superpixel-SLIC (simple linear iterative clustering) algorithm on CT images with COVID-19 for pulmonary parenchyma segmentation. To identify the pulmonary parenchyma, we described each superpixel cluster according to its position, grey intensity, second-order texture, and spatial-context-saliency features to classify by a tree random forest (TRF). Second, by applying the watershed segmentation to the mean-shift clusters, only pulmonary parenchyma segmentation-identified zones showed GGO and PI based on the description of each watershed cluster of its position, grey intensity, gradient entropy, second-order texture, Euclidean position to the border region of the PI zone, and global saliency features, after using TRF. Our classification results for pulmonary parenchyma identification on CT images with COVID-19 had a precision of over 92% and recall of over 92% on twofold cross validation. For GGO, the PI identification showed 96% precision and 96% recall on twofold cross validation.

CT radiomics facilitates more accurate diagnosis of COVID-19 pneumonia: compared with CO-RADS.

BACKGROUND: Limited data was available for rapid and accurate detection of COVID-19 using CT-based machine learning model. This study aimed to investigate the value of chest CT radiomics for diagnosing COVID-19 pneumonia compared with clinical model and COVID-19 reporting and data system (CO-RADS), and develop an open-source diagnostic tool with the constructed radiomics model. METHODS: This study enrolled 115 laboratory-confirmed COVID-19 and 435 non-COVID-19 pneumonia patients (training dataset, n = 379; validation dataset, n = 131; testing dataset, n = 40). Key radiomics features extracted from chest CT images were selected to build a radiomics signature using least absolute shrinkage and selection operator (LASSO) regression. Clinical and clinico-radiomics combined models were constructed. The combined model was further validated in the viral pneumonia cohort, and compared with performance of two radiologists using CO-RADS. The diagnostic performance was assessed by receiver operating characteristics curve (ROC) analysis, calibration curve, and decision curve analysis (DCA). RESULTS: Eight radiomics features and 5 clinical variables were selected to construct the combined radiomics model, which outperformed the clinical model in diagnosing COVID-19 pneumonia with an area under the ROC (AUC) of 0.98 and good calibration in the validation cohort. The combined model also performed better in distinguishing COVID-19 from other viral pneumonia with an AUC of 0.93 compared with 0.75 (P = 0.03) for clinical model, and 0.69 (P = 0.008) or 0.82 (P = 0.15) for two trained radiologists using CO-RADS. The sensitivity and specificity of the combined model can be achieved to 0.85 and 0.90. The DCA confirmed the clinical utility of the combined model. An easy-to-use open-source diagnostic tool was developed using the combined model. CONCLUSIONS: The combined radiomics model outperformed clinical model and CO-RADS for diagnosing COVID-19 pneumonia, which can facilitate more rapid and accurate detection.

Review on Diagnosis of COVID-19 from Chest CT Images Using Artificial Intelligence.

The COVID-19 diagnostic approach is mainly divided into two broad categories, a laboratory-based and chest radiography approach. The last few months have witnessed a rapid increase in the number of studies use artificial intelligence (AI) techniques to diagnose COVID-19 with chest computed tomography (CT). In this study, we review the diagnosis of COVID-19 by using chest CT toward AI. We searched ArXiv, MedRxiv, and Google Scholar using the terms "deep learning", "neural networks", "COVID-19", and "chest CT". At the time of writing (August 24, 2020), there have been nearly 100 studies and 30 studies among them were selected for this review. We categorized the studies based on the classification tasks: COVID-19/normal, COVID-19/non-COVID-19, COVID-19/non-COVID-19 pneumonia, and severity. The sensitivity, specificity, precision, accuracy, area under the curve, and F1 score results were reported as high as 100%, 100%, 99.62, 99.87%, 100%, and 99.5%, respectively. However, the presented results should be carefully compared due to the different degrees of difficulty of different classification tasks.

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

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.

Efficient and Effective Training of COVID-19 Classification Networks With Self-Supervised Dual-Track Learning to Rank.

Coronavirus Disease 2019 (COVID-19) has rapidly spread worldwide since first reported. Timely diagnosis of COVID-19 is crucial both for disease control and patient care. Non-contrast thoracic computed tomography (CT) has been identified as an effective tool for the diagnosis, yet the disease outbreak has placed tremendous pressure on radiologists for reading the exams and may potentially lead to fatigue-related mis-diagnosis. Reliable automatic classification algorithms can be really helpful; however, they usually require a considerable number of COVID-19 cases for training, which is difficult to acquire in a timely manner. Meanwhile, how to effectively utilize the existing archive of non-COVID-19 data (the negative samples) in the presence of severe class imbalance is another challenge. In addition, the sudden disease outbreak necessitates fast algorithm development. In this work, we propose a novel approach for effective and efficient training of COVID-19 classification networks using a small number of COVID-19 CT exams and an archive of negative samples. Concretely, a novel self-supervised learning method is proposed to extract features from the COVID-19 and negative samples. Then, two kinds of soft-labels ('difficulty' and 'diversity') are generated for the negative samples by computing the earth mover's distances between the features of the negative and COVID-19 samples, from which data 'values' of the negative samples can be assessed. A pre-set number of negative samples are selected accordingly and fed to the neural network for training. Experimental results show that our approach can achieve superior performance using about half of the negative samples, substantially reducing model training time.

Radiomics Analysis of Computed Tomography helps predict poor prognostic outcome in COVID-19.

Rationale: Given the rapid spread of COVID-19, an updated risk-stratify prognostic tool could help clinicians identify the high-risk patients with worse prognoses. We aimed to develop a non-invasive and easy-to-use prognostic signature by chest CT to individually predict poor outcome (death, need for mechanical ventilation, or intensive care unit admission) in patients with COVID-19. Methods: From November 29, 2019 to February 19, 2020, a total of 492 patients with COVID-19 from four centers were retrospectively collected. Since different durations from symptom onsets to the first CT scanning might affect the prognostic model, we designated the 492 patients into two groups: 1) the early-phase group: CT scans were performed within one week after symptom onset (0-6 days, n = 317); and 2) the late-phase group: CT scans were performed one week later after symptom onset (≥7 days, n = 175). In each group, we divided patients into the primary cohort (n = 212 in the early-phase group, n = 139 in the late-phase group) and the external independent validation cohort (n = 105 in the early-phase group, n = 36 in the late-phase group) according to the centers. We built two separate radiomics models in the two patient groups. Firstly, we proposed an automatic segmentation method to extract lung volume for radiomics feature extraction. Secondly, we applied several image preprocessing procedures to increase the reproducibility of the radiomics features: 1) applied a low-pass Gaussian filter before voxel resampling to prevent aliasing; 2) conducted ComBat to harmonize radiomics features per scanner; 3) tested the stability of the features in the radiomics signature by several image transformations, such as rotating, translating, and growing/shrinking. Thirdly, we used least absolute shrinkage and selection operator (LASSO) to build the radiomics signature (RadScore). Afterward, we conducted a Fine-Gray competing risk regression to build the clinical model and the clinic-radiomics signature (CrrScore). Finally, performances of the three prognostic signatures (clinical model, RadScore, and CrrScore) were estimated from the two aspects: 1) cumulative poor outcome probability prediction; 2) 28-day poor outcome prediction. We also did stratified analyses to explore the potential association between the CrrScore and the poor outcomes regarding different age, type, and comorbidity subgroups. Results: In the early-phase group, the CrrScore showed the best performance in estimating poor outcome (C-index = 0.850), and predicting the probability of 28-day poor outcome (AUC = 0.862). In the late-phase group, the RadScore alone achieved similar performance to the CrrScore in predicting poor outcome (C-index = 0.885), and 28-day poor outcome probability (AUC = 0.976). Moreover, the RadScore in both groups successfully stratified patients with COVID-19 into low- or high-RadScore groups with significantly different survival time in the training and validation cohorts (all P < 0.05). The CrrScore in both groups can also significantly stratify patients with different prognoses regarding different age, type, and comorbidities subgroups in the combined cohorts (all P < 0.05). Conclusions: This research proposed a non-invasive and quantitative prognostic tool for predicting poor outcome in patients with COVID-19 based on CT imaging. Taking the insufficient medical recourse into account, our study might suggest that the chest CT radiomics signature of COVID-19 is more effective and ideal to predict poor outcome in the late-phase COVID-19 patients. For the early-phase patients, integrating radiomics signature with clinical risk factors can achieve a more accurate prediction of individual poor prognostic outcome, which enables appropriate management and surveillance of COVID-19.