ABSTRACT
Background: The automatic analysis of medical images has the potential improve diagnostic accuracy while reducing the strain on clinicians. Current methods analyzing 3D-like imaging data, such as computerized tomography imaging, often treat each image slice as individual slices. This may not be able to appropriately model the relationship between slices. Methods: Our proposed method utilizes a mixed-effects model within the deep learning framework to model the relationship between slices. We externally validated this method on a data set taken from a different country and compared our results against other proposed methods. We evaluated the discrimination, calibration, and clinical usefulness of our model using a range of measures. Finally, we carried out a sensitivity analysis to demonstrate our methods robustness to noise and missing data. Results: In the external geographic validation set our model showed excellent performance with an AUROC of 0.930 (95%CI: 0.914, 0.947), with a sensitivity and specificity, PPV, and NPV of 0.778 (0.720, 0.828), 0.882 (0.853, 0.908), 0.744 (0.686, 0.797), and 0.900 (0.872, 0.924) at the 0.5 probability cut-off point. Our model also maintained good calibration in the external validation dataset, while other methods showed poor calibration. Conclusion: Deep learning can reduce stress on healthcare systems by automatically screening CT imaging for COVID-19. Our method showed improved generalizability in external validation compared to previous published methods. However, deep learning models must be robustly assessed using various performance measures and externally validated in each setting. In addition, best practice guidelines for developing and reporting predictive models are vital for the safe adoption of such models.
ABSTRACT
Coronavirus disease (COVID-19) has caused a worldwide pandemic, putting millions of people's health and lives in jeopardy. Detecting infected patients early on chest computed tomography (CT) is critical in combating COVID-19. Harnessing uncertainty-aware consensus-assisted multiple instance learning (UC-MIL), we propose to diagnose COVID-19 using a new bilateral adaptive graph-based (BA-GCN) model that can use both 2D and 3D discriminative information in 3D CT volumes with arbitrary number of slices. Given the importance of lung segmentation for this task, we have created the largest manual annotation dataset so far with 7,768 slices from COVID-19 patients, and have used it to train a 2D segmentation model to segment the lungs from individual slices and mask the lungs as the regions of interest for the subsequent analyses. We then used the UC-MIL model to estimate the uncertainty of each prediction and the consensus between multiple predictions on each CT slice to automatically select a fixed number of CT slices with reliable predictions for the subsequent model reasoning. Finally, we adaptively constructed a BA-GCN with vertices from different granularity levels (2D and 3D) to aggregate multi-level features for the final diagnosis with the benefits of the graph convolution network's superiority to tackle cross-granularity relationships. Experimental results on three largest COVID-19 CT datasets demonstrated that our model can produce reliable and accurate COVID-19 predictions using CT volumes with any number of slices, which outperforms existing approaches in terms of learning and generalisation ability. To promote reproducible research, we have made the datasets, including the manual annotations and cleaned CT dataset, as well as the implementation code, available at https://doi.org/10.5281/zenodo.6361963.
