Human-airway surface mesh smoothing based on graph convolutional neural networks.

BACKGROUND AND OBJECTIVE: A detailed representation of the airway geometry in the respiratory system is critical for predicting precise airflow and pressure behaviors in computed tomography (CT)-image-based computational fluid dynamics (CFD). The CT-image-based geometry often contains artifacts, noise, and discontinuities due to the so-called stair step effect. Hence, an advanced surface smoothing is necessary. The existing smoothing methods based on the Laplacian operator drastically shrink airway geometries, resulting in the loss of information related to smaller branches. This study aims to introduce an unsupervised airway-mesh-smoothing learning (AMSL) method that preserves the original geometry of the three-dimensional (3D) airway for accurate CT-image-based CFD simulations. METHOD: The AMSL method jointly trains two graph convolutional neural networks (GCNNs) defined on airway meshes to filter vertex positions and face normal vectors. In addition, it regularizes a combination of loss functions such as reproducibility, smoothness and consistency of vertex positions, and normal vectors. The AMSL adopts the concept of a deep mesh prior model, and it determines the self-similarity for mesh restoration without using a large dataset for training. Images of the airways of 20 subjects were smoothed by the AMSL method, and among them, the data of two subjects were used for the CFD simulations to assess the effect of airway smoothing on flow properties. RESULTS: In 18 of 20 benchmark problems, the proposed smoothing method delivered better results compared with the conventional or state-of-the-art deep learning methods. Unlike the traditional smoothing, the AMSL successfully constructed 20 smoothed airways with airway diameters that were consistent with the original CT images. Besides, CFD simulations with the airways obtained by the AMSL method showed much smaller pressure drop and wall shear stress than the results obtained by the traditional method. CONCLUSIONS: The airway model constructed by the AMSL method reproduces branch diameters accurately without any shrinkage, especially in the case of smaller airways. The accurate estimation of airway geometry using a smoothing method is critical for estimating flow properties in CFD simulations.

An unsupervised image registration method employing chest computed tomography images and deep neural networks.

BACKGROUND: Deformable image registration is crucial for multiple radiation therapy applications. Fast registration of computed tomography (CT) lung images is challenging because of the large and nonlinear deformation between inspiration and expiration. With advancements in deep learning techniques, learning-based registration methods are considered efficient alternatives to traditional methods in terms of accuracy and computational cost. METHOD: In this study, an unsupervised lung registration network (LRN) with cycle-consistent training is proposed to align two acquired CT-derived lung datasets during breath-holds at inspiratory and expiratory levels without utilizing any ground-truth registration results. Generally, the LRN model uses three loss functions: image similarity, regularization, and Jacobian determinant. Here, LRN was trained on the CT datasets of 705 subjects and tested using 10 pairs of public CT DIR-Lab datasets. Furthermore, to evaluate the effectiveness of the registration technique, target registration errors (TREs) of the LRN model were compared with those of the conventional algorithm (sum of squared tissue volume difference; SSTVD) and a state-of-the-art unsupervised registration method (VoxelMorph). RESULTS: The results showed that the LRN with an average TRE of 1.78 ± 1.56 mm outperformed VoxelMorph with an average TRE of 2.43 ± 2.43 mm, which is comparable to that of SSTVD with an average TRE of 1.66 ± 1.49 mm. In addition, estimating the displacement vector field without any folding voxel consumed less than 2 s, demonstrating the superiority of the learning-based method with respect to fiducial marker tracking and the overall soft tissue alignment with a nearly real-time speed. CONCLUSIONS: Therefore, this proposed method shows significant potential for use in time-sensitive pulmonary studies, such as lung motion tracking and image-guided surgery.

Classification of rotator cuff tears in ultrasound images using deep learning models.

Rotator cuff tears (RCTs) are one of the most common shoulder injuries, which are typically diagnosed using relatively expensive and time-consuming diagnostic imaging tests such as magnetic resonance imaging or computed tomography. Deep learning algorithms are increasingly used to analyze medical images, but they have not been used to identify RCTs with ultrasound images. The aim of this study is to develop an approach to automatically classify RCTs and provide visualization of tear location using ultrasound images and convolutional neural networks (CNNs). The proposed method was developed using transfer learning and fine-tuning with five pre-trained deep models (VGG19, InceptionV3, Xception, ResNet50, and DenseNet121). The Bayesian optimization method was also used to optimize hyperparameters of the CNN models. A total of 194 ultrasound images from Kosin University Gospel Hospital were used to train and test the CNN models by five-fold cross-validation. Among the five models, DenseNet121 demonstrated the best classification performance with 88.2% accuracy, 93.8% sensitivity, 83.6% specificity, and AUC score of 0.832. A gradient-weighted class activation mapping (Grad-CAM) highlighted the sensitive features in the learning process on ultrasound images. The proposed approach demonstrates the feasibility of using deep learning and ultrasound images to assist RCTs' diagnosis.

Structural and functional alterations of subjects with cement dust exposure: A longitudinal quantitative computed tomography-based study.

Cement dust exposure (CDE) can be a risk factor for pulmonary disease, causing changes in segmental airways and parenchymal lungs. This study investigates longitudinal alterations in quantitative computed tomography (CT)-based metrics due to CDE. We obtained CT-based airway structural and lung functional metrics from CDE subjects with baseline CT and follow-up CT scans performed three years later. From the CT, we extracted wall thickness (WT) and bifurcation angle (θ) at total lung capacity (TLC) and functional residual capacity (FRC), respectively. We also computed air volume (Vair), tissue volume (Vtissue), global lung shape, percentage of emphysema (Emph%), and more. Clinical measures were used to associate with CT-based metrics. Three years after their baseline, the pulmonary function tests of CDE subjects were similar or improved, but there were significant alterations in the CT-based structural and functional metrics. The follow-up CT scans showed changes in θ at most of the central airways; increased WT at the subgroup bronchi; smaller Vair at TLC at all except the right upper and lower lobes; smaller Vtissue at all lobes in TLC and FRC except for the upper lobes in FRC; smaller global lung shape; and greater Emph% at the right upper and lower lobes. CT-based structural and functional variables are more sensitive to the early identification of CDE subjects, while most clinical lung function changes were not noticeable. We speculate that the significant long-term changes in CT are uniquely observed in CDE subjects, different from smoking-induced structural changes.

A 3D-CNN model with CT-based parametric response mapping for classifying COPD subjects.

Chronic obstructive pulmonary disease (COPD) is a respiratory disorder involving abnormalities of lung parenchymal morphology with different severities. COPD is assessed by pulmonary-function tests and computed tomography-based approaches. We introduce a new classification method for COPD grouping based on deep learning and a parametric-response mapping (PRM) method. We extracted parenchymal functional variables of functional small airway disease percentage (fSAD%) and emphysema percentage (Emph%) with an image registration technique, being provided as input parameters of 3D convolutional neural network (CNN). The integrated 3D-CNN and PRM (3D-cPRM) achieved a classification accuracy of 89.3% and a sensitivity of 88.3% in five-fold cross-validation. The prediction accuracy of the proposed 3D-cPRM exceeded those of the 2D model and traditional 3D CNNs with the same neural network, and was comparable to that of 2D pretrained PRM models. We then applied a gradient-weighted class activation mapping (Grad-CAM) that highlights the key features in the CNN learning process. Most of the class-discriminative regions appeared in the upper and middle lobes of the lung, consistent with the regions of elevated fSAD% and Emph% in COPD subjects. The 3D-cPRM successfully represented the parenchymal abnormalities in COPD and matched the CT-based diagnosis of COPD.

Deep Learning Models for Predicting Severe Progression in COVID-19-Infected Patients: Retrospective Study.

BACKGROUND: Many COVID-19 patients rapidly progress to respiratory failure with a broad range of severities. Identification of high-risk cases is critical for early intervention. OBJECTIVE: The aim of this study is to develop deep learning models that can rapidly identify high-risk COVID-19 patients based on computed tomography (CT) images and clinical data. METHODS: We analyzed 297 COVID-19 patients from five hospitals in Daegu, South Korea. A mixed artificial convolutional neural network (ACNN) model, combining an artificial neural network for clinical data and a convolutional neural network for 3D CT imaging data, was developed to classify these cases as either high risk of severe progression (ie, event) or low risk (ie, event-free). RESULTS: Using the mixed ACNN model, we were able to obtain high classification performance using novel coronavirus pneumonia lesion images (ie, 93.9% accuracy, 80.8% sensitivity, 96.9% specificity, and 0.916 area under the curve [AUC] score) and lung segmentation images (ie, 94.3% accuracy, 74.7% sensitivity, 95.9% specificity, and 0.928 AUC score) for event versus event-free groups. CONCLUSIONS: Our study successfully differentiated high-risk cases among COVID-19 patients using imaging and clinical features. The developed model can be used as a predictive tool for interventions in aggressive therapies.