For deep learning networks used to segment organs at risk (OARs) in head and neck (H&N) cancers, the class-imbalance problem between small volume OARs and whole computed tomography (CT) images results in delineation with serious false-positives on irrelevant slices and unnecessary time-consuming calculations. To alleviate this problem, a slice classification model-facilitated 3D encoder–decoder network was developed and validated. In the developed two-step segmentation model, a slice classification model was firstly utilized to classify CT slices into six categories in the craniocaudal direction. Then the target categories for different OARs were pushed to the different 3D encoder–decoder segmentation networks, respectively. All the patients were divided into training (n = 120), validation (n = 30) and testing (n = 20) datasets. The average accuracy of the slice classification model was 95.99%. The Dice similarity coefficient and 95% Hausdorff distance, respectively, for each OAR were as follows: right eye (0.88 ± 0.03 and 1.57 ± 0.92 mm), left eye (0.89 ± 0.03 and 1.35 ± 0.43 mm), right optic nerve (0.72 ± 0.09 and 1.79 ± 1.01 mm), left optic nerve (0.73 ± 0.09 and 1.60 ± 0.71 mm), brainstem (0.87 ± 0.04 and 2.28 ± 0.99 mm), right temporal lobe (0.81 ± 0.12 and 3.28 ± 2.27 mm), left temporal lobe (0.82 ± 0.09 and 3.73 ± 2.08 mm), right temporomandibular joint (0.70 ± 0.13 and 1.79 ± 0.79 mm), left temporomandibular joint (0.70 ± 0.16 and 1.98 ± 1.48 mm), mandible (0.89 ± 0.02 and 1.66 ± 0.51 mm), right parotid (0.77 ± 0.07 and 7.30 ± 4.19 mm) and left parotid (0.71 ± 0.12 and 8.41 ± 4.84 mm). The total segmentation time was 40.13 s. The 3D encoder–decoder network facilitated by the slice classification model demonstrated superior performance in accuracy and efficiency in segmenting OARs in H&N CT images. This may significantly reduce the workload for radiation oncologists.

Objective The aim of this study was to explore the role of the AI system which was designed and developed based on the characteristics of COVID-19 CT images in the screening and evaluation of COVID-19. Methods The research team adopted an improved U-shaped neural network to segment lungs and pneumonia lesions in CT images through multilayer convolution iterations. Then the appropriate 159 cases were selected to establish and train the model, and Dice loss function and Adam optimizer were used for network training with the initial learning rate of 0.001. Finally, 39 cases (29 positive and 10 negative) were selected for the comparative test. Experimental group: an attending physician a and an associate chief physician a read the CT images to diagnose COVID-19 with the help of the AI system. Control group: an attending physician b and an associate chief physician b did the diagnosis only by their experience, without the help of the AI system. The time spent by each doctor in the diagnosis and their diagnostic results were recorded. Paired t -test, univariate ANOVA, chi-squared test, receiver operating characteristic curves, and logistic regression analysis were used for the statistical analysis. Results There was statistical significance in the time spent in the diagnosis of different groups ( P <0.05). For the group with the optimal diagnostic results, univariate and multivariate analyses both suggested no significant correlation for all variables, and thus it might be the assistance of the AI system, the epidemiological history and other factors that played an important role. Conclusion The AI system developed by us, which was created due to COVID-19, had certain clinical practicability and was worth popularizing.

<b><i>Background:</i></b> Due to the similar symptoms of upper airway obstruction to asthma, misdiagnosis is common. Spirometry is a cost-effective screening test for upper airway obstruction and its characteristic patterns involving fixed, variable intrathoracic and extrathoracic lesions. We aimed to develop a deep learning model to detect upper airway obstruction patterns and compared its performance with that of lung function clinicians. <b><i>Methods:</i></b> Spirometry records were reviewed to detect the possible condition of airway stenosis. Then they were confirmed by the gold standard (e.g., computed tomography, endoscopy, or clinic diagnosis of upper airway obstruction). Images and indices derived from flow-volume curves were used for training and testing the model. Clinicians determined cases using spirometry records from the test set. The deep learning model evaluated the same data. <b><i>Results:</i></b> Of 45,831 patients’ spirometry records, 564 subjects with curves suggesting upper airway obstruction, after verified by the gold standard, 351 patients were confirmed. These cases and another 200 cases without airway stenosis were used as the training and testing sets. 432 clinicians evaluated 20 cases of each of the three patterns and 20 no airway stenosis cases (<i>n</i> = 80). They assigned an accuracy of 41.2% (±15.4) (interquartile range: 27.5–52.5%), with poor agreements (κ = 0.12). For the same cases, the model generated a correct detection of 81.3% (<i>p</i> &#x3c; 0.0001). <b><i>Conclusions:</i></b> Deep learning could detect upper airway obstruction patterns from other classic patterns of ventilatory defects with high accuracy, whereas clinicians presented marked errors and variabilities. The model may serve as a support tool to enhance clinicians’ correct diagnosis of upper airway obstruction using spirometry.

No abstract