The development of medical assisting tools based on artificial intelligence advances is essential in the global fight against COVID-19 outbreak and the future of medical systems. In this study, we introduce ai-corona, a radiologist-assistant deep learning framework for COVID-19 infection diagnosis using chest CT scans. Our framework incorporates an EfficientNetB3-based feature extractor. We employed three datasets; the CC-CCII set, the MasihDaneshvari Hospital (MDH) cohort, and the MosMedData cohort. Overall, these datasets constitute 7184 scans from 5693 subjects and include the COVID-19, non-COVID abnormal (NCA), common pneumonia (CP), non-pneumonia, and Normal classes. We evaluate ai-corona on test sets from the CC-CCII set, MDH cohort, and the entirety of the MosMedData cohort, for which it gained AUC scores of 0.997, 0.989, and 0.954, respectively. Our results indicates ai-corona outperforms all the alternative models. Lastly, our framework’s diagnosis capabilities were evaluated as assistant to several experts. Accordingly, We observed an increase in both speed and accuracy of expert diagnosis when incorporating ai-corona’s assistance.

Background: With the global outbreak of COVID-19 epidemic since early 2020, there has been considerable attention on CT-based diagnosis as an effective and reliable method. Recently, the advent of deep learning in medical diagnosis has been well proven. Convolutional Neural Networks (CNN) can be used to detect the COVID-19 infection imaging features in a chest CT scan. We introduce ai-corona, a radiologist-assistant deep learning framework for COVID-19 infection diagnosis using the chest CT scans. Method: Our dataset comprises 2121 cases of axial spiral chest CT scans in three classes; COVID-19 abnormal, non COVID-19 abnormal, and normal, from which 1764 cases were used for training and 357 cases for validation. The training set was annotated using the reports of two experienced radiologists. The COVID-19 abnormal class validation set was annotated using the general consensus of a collective of criteria that indicate COVID-19 infection. Moreover, the validation sets for the non COVID-19 abnormal and the normal classes were annotated by a different experienced radiologist. ai-corona constitutes a CNN-based feature extractor conjoined with an average pooling and a fully-connected layer to classify a given chest CT scan into the three aforementioned classes. Results: We compare the diagnosis performance of ai-corona, radiologists, and model-assisted radiologists for six combinations of distinguishing between the three mentioned classes, including COVID-19 abnormal vs. others, COVID-19 abnormal vs. normal, COVID-19 abnormal vs. non COVID-19 abnormal, non COVID-19 abnormal vs. others, normal vs. others, and normal vs. abnormal. ai-corona achieves an AUC score of 0.989 (95% CI: 0.984, 0.994), 0.997 (95% CI: 0.995, 0.999), 0.986 (95% CI: 0.981, 0.991), 0.959 (95% CI: 0.944, 0.974), 0.978 (95% CI: 0.968, 0.988), and 0.961 (95% CI: 0.951, 0.971) in each combination, respectively. By employing Bayesian statistics to calculate the accuracies at a 95% confidence interval, ai-corona surpasses the radiologists in distinguishing between the COVID-19 abnormal class and the other two classes (especially the non COVID-19 abnormal class). Our results show that radiologists' diagnosis performance improves when incorporating ai-corona's prediction. In addition, we also show that RT-PCR's diagnosis has a much lower sensitivity compared to all the other methods.

No abstract

With the onset of the COVID-19 pandemic, quantifying the condition of positively diagnosed patients is of paramount importance. Chest CT scans can be used to measure the severity of a lung infection and the isolate involvement sites in order to increase awareness of a patient's disease progression. In this work, we developed a deep learning framework for lung infection severity prediction. To this end, we collected a dataset of 232 chest CT scans and involved two public datasets with an additional 59 scans for our model's training and used two external test sets with 21 scans for evaluation. On an input chest Computer Tomography (CT) scan, our framework, in parallel, performs a lung lobe segmentation utilizing a pre-trained model and infection segmentation using three distinct trained SE-ResNet18 based U-Net models, one for each of the axial, coronal, and sagittal views. By having the lobe and infection segmentation masks, we calculate the infection severity percentage in each lobe and classify that percentage into 6 categories of infection severity score using a k-nearest neighbors (k-NN) model. The lobe segmentation model achieved a Dice Similarity Score (DSC) in the range of [0.918, 0.981] for different lung lobes and our infection segmentation models gained DSC scores of 0.7254 and 0.7105 on our two test sets, respectfully. Similarly, two resident radiologists were assigned the same infection segmentation tasks, for which they obtained a DSC score of 0.7281 and 0.6693 on the two test sets. At last, performance on infection severity score over the entire test datasets was calculated, for which the framework's resulted in a Mean Absolute Error (MAE) of 0.505 ± 0.029, while the resident radiologists' was 0.571 ± 0.039.

No abstract