ABSTRACT
The global spread of COVID-19 seriously endangers human health and even lives. By predicting patients' individualized disease development and further performing intervention in time, we may rationalize scarce medical resources and reduce mortality. Based on 1337 multi- stage ([≥]3) high-resolution chest computed tomography (CT) images of 417 infected patients from three centers in the epidemic area, we proposed a random forest + cellular automata (RF+CA) model to forecast voxel-level lesion development of patients with COVID-19. The model showed a promising prediction performance (Dice similarity coefficient [DSC] = 71.1%, Kappa coefficient = 0.612, Figure of Merit [FoM] = 0.257, positional accuracy [PA] = 3.63) on the multicenter dataset. Using this model, multiple driving factors for the development of lesions were determined, such as distance to various interstitials in the lung, distance to the pleura, etc. The driving processes of these driving factors were further dissected and explained in depth from the perspective of pathophysiology, to explore the mechanism of individualized development of COVID-19 disease. The complete codes of the forecast system are available at https://github.com/keyunj/VVForecast_covid19.
Subject(s)
COVID-19ABSTRACT
Segmentation of infected areas in chest CT volumes is of great significance for further diagnosis and treatment of COVID-19 patients. Due to the complex shapes and varied appearances of lesions, a large number of voxel-level labeled samples are generally required to train a lesion segmentation network, which is a main bottleneck for developing deep learning based medical image segmentation algorithms. In this paper, we propose a weakly-supervised lesion segmentation framework by embedding the Generative Adversarial training process into the Segmentation Network, which is called GASNet. GASNet is optimized to segment the lesion areas of a COVID-19 CT by the segmenter, and to replace the abnormal appearance with a generated normal appearance by the generator, so that the restored CT volumes are indistinguishable from healthy CT volumes by the discriminator. GASNet is supervised by chest CT volumes of many healthy and COVID-19 subjects without voxel-level annotations. Experiments on three public databases show that when using as few as one voxel-level labeled sample, the performance of GASNet is comparable to fully-supervised segmentation algorithms trained on dozens of voxel-level labeled samples.
Subject(s)
COVID-19ABSTRACT
Early detection of COVID-19 based on chest CT will enable timely treatment of patients and help control the spread of the disease. With rapid spreading of COVID-19 in many countries, however, CT volumes of suspicious patients are increasing at a speed much faster than the availability of human experts. We proposed an artificial intelligence (AI) system for fast COVID-19 detection and performed extensive statistical analysis of CTs of COVID-19 based on the AI system. We developed and evaluated our system on a large dataset with more than 10 thousand CT volumes from COVID-19, influenza-A/B, non-viral community acquired pneumonia (CAP) and non-pneumonia subjects. In such a difficult multi-class diagnosis task, our deep convolutional neural network-based system is able to achieve an area under the receiver operating characteristic curve (AUC) of 97.17%, a sensitivity of 90.19%, and a specificity of 95.76% for COVID-19 on internal test cohort of 3,203 scans and AUC of 97.77% on the publicly available CC-CCII database with 1,943 test samples. In a reader study involving five radiologists, the AI system outperforms all of radiologists in more challenging tasks at a speed of two orders of magnitude above them. Diagnosis performance of chest x-ray (CXR) is compared. Detailed interpretation of deep network is also performed to relate AI results with CT findings. The code is available at https://github.com/ChenWWWeixiang/diagnosis_covid19.