| Literature DB >> 36237723 |
Seung-Jin Yoo, Jin Mo Goo, Soon Ho Yoon.
Abstract
Coronavirus disease (COVID-19) has threatened public health as a global pandemic. Chest CT and radiography are crucial in managing COVID-19 in addition to reverse transcription-polymerase chain reaction, which is the gold standard for COVID-19 diagnosis. This is a review of the current status of the use of chest CT and radiography in COVID-19 diagnosis and management and an introduction of early representative studies on the application of artificial intelligence to chest CT and radiography. The authors also share their experiences to provide insights into the future value of artificial intelligence. CopyrightsEntities:
Year: 2020 PMID: 36237723 PMCID: PMC9431829 DOI: 10.3348/jksr.2020.0138
Source DB: PubMed Journal: Taehan Yongsang Uihakhoe Chi ISSN: 1738-2637
Summary of Representative Published Studies on CT Artificial Intelligence for COVID-19
| References | Aim | Consecutive Positive/Negative Data Collection during the Pandemic | Positive Data (COVID-19) | Negative Data | Deep Neural Network | Internal Validation Dataset (%) | External Validation Dataset (%) | Human Interaction (%) | Code or Program Availability | ||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Sen. | Spe. | Sen. | Spe. | ||||||||
| Li et al. ( | Diagnosis | No | 1296 scans | 1735 scans, CAP; 1325 scans, non-pneumonia | COVNex based on ResNet | 90 | 87 | Yes | |||
| Wang et al. ( | Diagnosis Prognosis | No | 924 scans | 342 scans, other pneumonia | DenseNet-like structure | 79 | 90 | 79–80 | 77–81 | Yes | |
| Mei et al. ( | Diagnosis | Yes | 419 scans | 486 scans, negative for RT-PCR | 2 CNNs and ML classifiers | 84 | 83 | 79 → 88 | Yes | ||
| Bai et al. ( | Diagnosis | No | 521 scans | 665 scans, other pneumonia | EfficientNet | 96 | 95 | 89 | 86 | Yes | |
| Ardakani et al. ( | Diagnosis | No | 108 scans | 86 scans, other pneumonia | ResNet | 100 | 99 | No | |||
| Song et al. ( | Diagnosis | Yes | 83 scans | 83 scans negative for RT-PCR | BigBiGAN | 92 | 91 | 80 | 75 | 77 → 85; | Yes |
| Wu et al. ( | Diagnosis | Unknown | 368 scans | 127 scans, other pneumonia | ResNet | 81 | 65 | No | |||
| Jaiswal et al. ( | Diagnosis | Unknown | 1262 slices | 1230 slices, negative for RT-PCR | DenseNet | 96 | 96 | No | |||
| Kang et al. ( | Diagnosis | No | 1495 scans | 1027 scans, CAP | Latent-representation regressor with FCNN | 97 | 93 | No | |||
| Wang et al. ( | Diagnosis | No | 1315 scans | 2406 scans, viral ILD; 936 scans, normal | ResNet | 88 | 96 | No | |||
| Ouyang et al. ( | Diagnosis | Unknown | 3389 images | 1593 images, CAP | 3D CNN | 87 | 90 | No | |||
| Wang et al. ( | Diagnosis | Yes | 313 scans | 229 scans negative for RT-PCR | 3D CNN | 84 | 98 | Yes | |||
| Han et al. ( | Diagnosis | No | 230 scans | 100 scans, other pneumonia; 130 scans, without pneumonia | 3D CNN | 91 | 96 | Yes | |||
| Ko et al. ( | Diagnosis | No | 1194 images | 1357 images, other pneumonia; 998 images, normal; 444 images, lung cancer | 2D CNN | 99 | 100 | Accuracy of 97 | No | ||
| Ni et al. ( | Diagnosis | Unknown | 3854 scans | 6871 scans, other pneumonia; 8566 scans, normal | MVP net | 100 | 25 | 89 → 94 | No | ||
| Hu et al. ( | Diagnosis | Unknown | 150 scans | 150 scans, other pneumonia; 150 scans, non-pneumonia | CNN | 89 | 88 | No | |||
| Wang et al. ( | Segmentation | Unknown | 558 scans | None | 2D U-Net | Dice, 81 | Yes | ||||
| Zhou et al. ( | Segmentation | Unknown | 201 scans | None | 2.5D U-Net | Dice, 78 | Yes | ||||
CAP = community-acquired pneumonia, CNN = convolutional neural network, COVID-19 = coronavirus disease 2019, CPA = community acquired pneumonia, D = dimensional, GAN = generative adversarial network, ILD = interstitial lung disease, MVP = Multi-View FPN with Position-aware attention, RT-PCR = reverse transcription-polymerase chain reaction, Sen = sensitivity, Spe = specificity
Summary of Representative Published Studies on Chest Radiography Artificial Intelligence for COVID-19
| References | Aim | Consecutive Positive/Negative Data Collection during the Pandemic | Positive Data (COVID-19) | Negative Data | Deep Neural Network | Internal Validation Dataset (%) | External Validation Dataset (%) | Human Interaction | Code or Program Availability | ||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Sen. | Spe. | Sen. | Spe. | ||||||||
| Das et al. ( | Diagnosis | No | 0.2 | 4280 images, other viral and bacterial pneumonia; 400 images, tuberculosis; 2003 images, normal | Inception Net V3 | 95 | 99 | No | |||
| Waheed et al. ( | Diagnosis | No | 403 images | 721 images, normal | VGG-16 | 69 | 95 | No | |||
| Additional 1669 synthetic images* | Additional 1399 synthetic images, normal* | VGG-16 | 90 | 97 | No | ||||||
| Oh et al. ( | Diagnosis | No | 180 images | 20 images, other viral pneumonia; 54 images, bacterial pneumonia; 57 images, tuberculosis; 191 images, normal | ResNet-18 | 93† | 89–96† | No | |||
| Altan & Karasu ( | Diagnosis | No | 219 images and 2441 synthetic images‡ | 1345 images, viral pneumonia; 1315 synthetic images, viral pneumonia; 1341 images, normal; 1319 synthetic images, normal‡ | EfficientNet-B0 | Accuracy of 95–99 | 96–99 | No | |||
| Rahimzadeh & Attar ( | Diagnosis | No | 180 images | 6054 images, pneumonia; 8851 images, normal | Concatenated Network of Xception and ResNet50V2 | 81 | 99 | Yes | |||
| Panwar et al. ( | Diagnosis | No | 142 images | 142 images, normal | nCOVnet based on VGG-16 | 98 | 79 | No | |||
| Brunese et al. ( | Diagnosis | No | 250 images | 2753 images, other pulmonary disease; 3520 images, normal | VGG-16 | 87 | 98 | No | |||
*The auxiliary classifier generative adversarial network (ACGAN) - a base model called CovidGAN was used for synthetic image augmentation and the discriminator model.
†Sensitivity and specificity for the diagnosis of viral pneumonia, including COVID-19 pneumonia, using chest X-ray images.
‡Data synthesis was performed using image processing techniques such as rotations, image resizing, and adding new pixel blocks in the horizontal and vertical directions.
COVID-19 = coronavirus disease 2019, Sen = sensitivity, Spe = specificity
Fig. 1A 26-year-old male patient diagnosed with COVID-19.
A. Chest CT scan shows peripheral consolidations and ground-glass opacities in the right middle, right lower, and left lower lobes, which is consistent with COVID-19 pneumonia.
B, C. A ground truth mask (B) created by the radiologist and an automatically segmented mask (C) created by a 2D U-Net match almost perfectly. The Dice similarity coefficient was 84.5%.
COVID-19 = coronavirus disease 2019