Search

BACKGROUND: Pulmonary granulomatous nodules (GN) with spiculation or lobulation have a similar morphological appearance to solid lung adenocarcinoma (SADC) under computed tomography (CT). However, these two kinds of solid pulmonary nodules (SPN) have different malignancies and are sometimes misdiagnosed. OBJECTIVE: This study aims to predict malignancies of SPNs by a deep learning model automatically. METHODS: A chimeric label with self-supervised learning (CLSSL) is proposed to pre-train a ResNet-based network (CLSSL-ResNet) for distinguishing isolated atypical GN from SADC in CT images. The malignancy, rotation, and morphology labels are integrated into a chimeric label and utilized to pre-train a ResNet50. The pre-trained ResNet50 is then transferred and fine-tuned to predict the malignancy of SPN. Two image datasets of 428 subjects (Dataset1, 307; Dataset2, 121) from different hospitals are collected. Dataset1 is divided into training, validation, and test data by a ratio of 7:1:2 to develop the model. Dataset2 is utilized as an external validation dataset. RESULTS: CLSSL-ResNet achieves an area under the ROC curve (AUC) of 0.944 and an accuracy (ACC) of 91.3%, which was much higher than that of the consensus of two experienced chest radiologists (77.3%). CLSSL-ResNet also outperforms other self-supervised learning models and many counterparts of other backbone networks. In Dataset2, AUC and ACC of CLSSL-ResNet are 0.923 and 89.3%, respectively. Additionally, the ablation experiment result indicates higher efficiency of the chimeric label. CONCLUSION: CLSSL with morphology labels can increase the ability of feature representation by deep networks. As a non-invasive method, CLSSL-ResNet can distinguish GN from SADC via CT images and may support clinical diagnoses after further validation. [ABSTRACT FROM AUTHOR]

Objectives: Few studies have explored the use of spectral dual-layer detector-based computed tomography (SDCT) parameters, thymidine kinase-1 (TK1), and tumor abnormal protein (TAP) for the detection of ground-glass nodules (GGNs). Therefore, we aimed to evaluate the quantitative and qualitative parameters generated from SDCT for predicting the pathological subtypes of GGN-featured lung adenocarcinoma combined with TK1 and TAP. Material and Methods: Between July 2021 and September 2022, 238 patients with GGNs were retrospectively enrolled in this study. SDCT and tests for TK1 and TAP were performed preoperatively, and the lesions were divided into glandular precursor lesions (PGL), minimally invasive adenocarcinoma (MIA), and invasive adenocarcinoma (IAC), according to the pathological results. A receiver operating characteristic (ROC) curve was used to compare the diagnostic performance of these parameters. Multivariate logistic regression analysis was performed to construct a joint diagnostic model and create a nomogram. Results: This study included 238 GGNs, including 41 atypical adenomatous hyperplasias (AAH), 62 adenocarcinomas in situ (AIS), 49 MIA, and 86 IAC, with a high proportion of women, non-smokers, and pure ground-glass nodule (pGGN). CT100 keV (a/v), electronic density (EDW) (a/v), Daverage, Dsolid, TK1, and TAP of MIA and IAC were higher than those of PGL. The effective atomic number (Zeff (a/v)) was lower in MIA and IAC than in PGL (all p < 0.05). Logistic regression analysis showed that Zeff (a), EDW (a), TK1, Daverage, and internal bronchial morphology were crucial factors in predicting the aggressiveness of GGN. Zeff (a) had the highest diagnostic performance with an area under the ROC curve (AUC) = 0.896, followed by EDW (a) (AUC = 0.838) and CT100 keVa (AUC = 0.819). The diagnostic model and nomogram constructed using these five parameters (Zeff (a) + EDW (a) + CT100 keVa + Daverage + TK1) had an AUC = 0.933, which was higher than the individual parameters (p < 0.05). Conclusions: Multiple quantitative and functional parameters can be selected based on SDCT, especially Zeff (a) and EDW (a), which have high sensitivity and specificity for predicting GGNs' invasiveness. Additionally, the combination of TK1 can further improve diagnostic performance, and using a nomogram is helpful for individualized predictions. [ABSTRACT FROM AUTHOR]

The individual prognosis of chemotherapy is quite different in non-small cell lung cancer (NSCLC). There is an urgent need to precisely predict and assess the treatment response. To develop a deep multiple-instance learning (DMIL) based model for predicting chemotherapy response in NSCLC in pretreatment CT images. Two datasets of NSCLC patients treated with chemotherapy as the first-line treatment were collected from two hospitals. Dataset 1 (163 response and 138 nonresponse) was used to train, validate, and test the DMIL model and dataset 2 (22 response and 20 nonresponse) was used as the external validation cohort. Five backbone networks in the feature extraction module and three pooling methods were compared. The DMIL with a pre-trained VGG16 backbone and an attention mechanism pooling performed the best, with an accuracy of 0.883 and area under the curve (AUC) of 0.982 on Dataset 1. While using max pooling and convolutional pooling, the AUC was 0.958 and 0.931, respectively. In Dataset 2, the best DMIL model produced an accuracy of 0.833 and AUC of 0.940. Deep learning models based on the MIL can predict chemotherapy response in NSCLC using pretreatment CT images and the pre-trained VGG16 with attention mechanism pooling yielded better predictions. [ABSTRACT FROM AUTHOR]

The heterogeneity and complexity of non-small cell lung cancer (NSCLC) tumors mean that NSCLC patients at the same stage can have different chemotherapy prognoses. Accurate predictive models could recognize NSCLC patients likely to respond to chemotherapy so that they can be given personalized and effective treatment. We propose to identify predictive imaging biomarkers from pre-treatment CT images and construct a radiomic model that can predict the chemotherapy response in NSCLC. This single-center cohort study included 280 NSCLC patients who received first-line chemotherapy treatment. Non-contrast CT images were taken before and after the chemotherapy, and clinical information were collected. Based on the Response Evaluation Criteria in Solid Tumors and clinical criteria, the responses were classified into two categories: response (n = 145) and progression (n = 135), then all data were divided into two cohorts: training cohort (224 patients) and independent test cohort (56 patients). In total, 1629 features characterizing the tumor phenotype were extracted from a cube containing the tumor lesion cropped from the pre-chemotherapy CT images. After dimensionality reduction, predictive models of the chemotherapy response of NSCLC with different feature selection methods and different machine-learning classifiers (support vector machine, random forest, and logistic regression) were constructed. For the independent test cohort, the predictive model based on a random-forest classifier with 20 radiomic features achieved the best performance, with an accuracy of 85.7% and an area under the receiver operating characteristic curve of 0.941 (95% confidence interval, 0.898–0.982). Of the 20 selected features, four were first-order statistics of image intensity and the others were texture features. For nine features, there were significant differences between the response and progression groups (p < 0.001). In the response group, three features, indicating heterogeneity, were overrepresented and one feature indicating homogeneity was underrepresented. The proposed radiomic model with pre-chemotherapy CT features can predict the chemotherapy response of patients with non-small cell lung cancer. This radiomic model can help to stratify patients with NSCLC, thereby offering the prospect of better treatment. [ABSTRACT FROM AUTHOR]

Dispositions of inhalable particles in the human respiratory tract trigger and exacerbate airway inflammatory diseases. However, the particle deposition (PD) in airway of subjects with tracheal bronchus (TB) and chronic obstructive pulmonary diseases (COPD) is unknown. We therefore propose to clarify the disrupted PD associated with TB and COPD using the computational fluid dynamics (CFD) simulation. Totally nine airway tree models are included. Six are extracted from CT images of different individuals (two with TB, two with COPD, and two healthy controls (HC)). The others are the artificially modified models (AMMs) generated by the virtual lesion. Specifically, they are constructed through artificially adding a tracheal bronchus or a stenosis on one HC model. The deposition efficiency (DE) and deposition fraction (DF) in these models are obtained by the Euler-Lagrange approach, analyzed, and compared across models, locations, and particle sizes (0.1-10.0 micrometers). It is found that the PD in models with TB and COPD has been disrupted by the geometrical changes and followed airflow alternations. DE of the tracheal bronchus is higher for TB models. For COPD, the stenosis location determines the effects on DE and DF. Higher DF at the trachea is observed in TB1, TB2, and COPD2 models. DE increases with the particle size, and DE of the terminal bronchi is higher than that of central regions. Combined with AMMs, the CFD simulation using realistic airway models demonstrates disruptions of DP. The methods and findings might help understand the etiology of pulmonary diseases and improve the efficacy of inhaled medicines. [ABSTRACT FROM AUTHOR]

Purpose: MRI is increasingly being used for radiotherapy planning, simulation, and in-treatmentroom motion monitoring. To provide more detailed temporal and spatial MR data for these tasks, we have recently developed a novel self-gated (SG) MRI technique with advantage of k-space phase sorting, high isotropic spatial resolution, and high temporal resolution. The current work describes the validation of this 4D-MRI technique using a MRI- and CT-compatible respiratory motion phantom and comparison to 4D-CT. Methods: The 4D-MRI sequence is based on a spoiled gradient echo-based 3D projection reconstruction sequence with self-gating for 4D-MRI at 3 T. Respiratory phase is resolved by using SG k-space lines as the motion surrogate. 4D-MRI images are reconstructed into ten temporal bins with spatial resolution 1.56×1.56×1.56 mm³. A MRI-CT compatible phantom was designed to validate the performance of the 4D-MRI sequence and 4D-CT imaging. A spherical target (diameter 23 mm, volume 6.37 ml) filled with high-concentration gadolinium (Gd) gel is embedded into a plastic box (35×40×63 mm³) and stabilized with low-concentration Gd gel. The phantom, driven by an air pump, is able to produce human-type breathing patterns between 4 and 30 respiratory cycles/min. 4D-CT of the phantom has been acquired in cine mode, and reconstructed into ten phases with slice thickness 1.25 mm. The 4D images sets were imported into a treatment planning software for target contouring. The geometrical accuracy of the 4D MRI and CT images has been quantified using target volume, flattening, and eccentricity. The target motion was measured by tracking the centroids of the spheres in each individual phase. Motion ground-truth was obtained from input signals and real-time video recordings. Results: The dynamic phantom has been operated in four respiratory rate (RR) settings, 6, 10, 15, and 20/min, and was scanned with 4D-MRI and 4D-CT. 4D-CT images have target-stretching, partialmissing, and other motion artifacts in various phases, whereas the 4D-MRI images are visually free of those artifacts. Volume percentage difference for the 6.37 ml target ranged from 5.3%±4.3% to 10.3%±5.9% for 4D-CT, and 1.47±0.52 to 2.12±1.60 for 4D-MRI. With an increase of respiratory rate, the target volumetric and geometric deviations increase for 4D-CT images while remaining stable for the 4D-MRI images. Target motion amplitude errors at different RRs were measured with a range of 0.66-1.25 mm for 4D-CT and 0.2-0.42 mm for 4D-MRI. The results of Mann-Whitney tests indicated that 4D-MRI significantly outperforms 4D-CT in phase-based target volumetric (p = 0.027) and geometric (p < 0.001) measures. Both modalities achieve equivalent accuracy in measuring motion amplitude (p = 0.828). Conclusions: The k-space self-gated 4D-MRI technique provides a robust method for accurately imaging phase-based target motion and geometry. Compared to 4D-CT, the current 4D-MRI technique demonstrates superior spatiotemporal resolution, and robust resistance to motion artifacts caused by fast target motion and irregular breathing patterns. The technique can be used extensively in abdominal targeting, motion gating, and toward implementing MRI-based adaptive radiotherapy. [ABSTRACT FROM AUTHOR]

Objective: We evaluated 3-D computed tomography angiography (3-D CTA) in the diagnosis of the nutcracker phenomenon, and its significance in postoperative follow up. Patients and Methods: Three-dimensional CTA was used to compare the anatomical relations of the left renal vein with the aorta and the superior mesenteric artery in patients with the nutcracker phenomenon and in a control group. Four patients with the nutcracker phenomenon received a surgical procedure of the transposition of the left renal vein. The 3-D CTA was used for all patients during postoperation follow-up testing. Results: The 3-D CTA showed a compression of the left renal vein between the aorta and the superior mesenteric artery (SMA) and the abnormal acute angle between them. The angles and distances between the SMA and the aorta were 39.3 ± 4.3° and 3.1 ± 0.2 mm in the patient groups and 90 ± 10° and 12 ± 1.8 mm in the control groups, respectively. Differences in angles and distances were statistically significant between the two groups ( P < 0.05). Surgical transposition of the left renal vein was performed successfully. Postoperative 3-D CTA revealed the distance between the SMA and the aorta was nearly normal. Conclusion: The reconstruction imaging of the renal vein by means of 3-D CTA revealed that unusual hematuria was due to compression of the left renal vein; therefore it may be a useful alternative imaging technique instead of conventional examinations. The non-invasive 3-D CTA may be a useful tool in the diagnosis of the nutcracker phenomenon and follow-up testing. [ABSTRACT FROM AUTHOR]

Background: Lung segmentation constitutes a critical procedure for any clinical-decision supporting system aimed to improve the early diagnosis and treatment of lung diseases. Abnormal lungs mainly include lung parenchyma with commonalities on CT images across subjects, diseases and CT scanners, and lung lesions presenting various appearances. Segmentation of lung parenchyma can help locate and analyze the neighboring lesions, but is not well studied in the framework of machine learning.Methods: We proposed to segment lung parenchyma using a convolutional neural network (CNN) model. To reduce the workload of manually preparing the dataset for training the CNN, one clustering algorithm based method is proposed firstly. Specifically, after splitting CT slices into image patches, the k-means clustering algorithm with two categories is performed twice using the mean and minimum intensity of image patch, respectively. A cross-shaped verification, a volume intersection, a connected component analysis and a patch expansion are followed to generate final dataset. Secondly, we design a CNN architecture consisting of only one convolutional layer with six kernels, followed by one maximum pooling layer and two fully connected layers. Using the generated dataset, a variety of CNN models are trained and optimized, and their performances are evaluated by eightfold cross-validation. A separate validation experiment is further conducted using a dataset of 201 subjects (4.62 billion patches) with lung cancer or chronic obstructive pulmonary disease, scanned by CT or PET/CT. The segmentation results by our method are compared with those yielded by manual segmentation and some available methods.Results: A total of 121,728 patches are generated to train and validate the CNN models. After the parameter optimization, our CNN model achieves an average F-score of 0.9917 and an area of curve up to 0.9991 for classification of lung parenchyma and non-lung-parenchyma. The obtain model can segment the lung parenchyma accurately for 201 subjects with heterogeneous lung diseases and CT scanners. The overlap ratio between the manual segmentation and the one by our method reaches 0.96.Conclusions: The results demonstrated that the proposed clustering algorithm based method can generate the training dataset for CNN models. The obtained CNN model can segment lung parenchyma with very satisfactory performance and have the potential to locate and analyze lung lesions. [ABSTRACT FROM AUTHOR]

Tracheal Bronchus (TB) is a rare congenital anomaly characterized by the presence of an abnormal bronchus originating from the trachea or main bronchi and directed toward the upper lobe. The airflow pattern in tracheobronchial trees of TB subjects is critical, but has not been systemically studied. This study proposes to simulate the airflow using CT image based models and the computational fluid dynamics (CFD) method. Six TB subjects and three health controls (HC) are included. After the geometric model of tracheobronchial tree is extracted from CT images, the spatial distribution of velocity, wall pressure, wall shear stress (WSS) is obtained through CFD simulation, and the lobar distribution of air, flow pattern and global pressure drop are investigated. Compared with HC subjects, the main bronchus angle of TB subjects and the variation of volume are large, while the cross-sectional growth rate is small. High airflow velocity, wall pressure, and WSS are observed locally at the tracheal bronchus, but the global patterns of these measures are still similar to those of HC. The ratio of airflow into the tracheal bronchus accounts for 6.6-15.6% of the inhaled airflow, decreasing the ratio to the right upper lobe from 15.7-21.4% (HC) to 4.9-13.6%. The air into tracheal bronchus originates from the right dorsal near-wall region of the trachea. Tracheal bronchus does not change the global pressure drop which is dependent on multiple variables. Though the tracheobronchial trees of TB subjects present individualized features, several commonalities on the structural and airflow characteristics can be revealed. The observed local alternations might provide new insight into the reason of recurrent local infections, cough and acute respiratory distress related to TB. [ABSTRACT FROM AUTHOR]

• The lymphoma aggressiveness prediction is formulated as a few-shot MIL problem. • Deep-learning-based and radiomics features from PET and CT are combined. • Our proposed bag-level few-shot learning framework is utilized for tackling the issue of limited NHL samples and incomplete annotations. Patients with aggressive non-Hodgkin lymphoma (NHL) undergo distinct therapy strategies compared with indolent NHL patients. However, it is challenging to estimate NHL aggressiveness based on visual inspection of positron emission tomography (PET) or computed tomography (CT) images. Since diffuse large B-cell lymphoma (DLBCL) and Follicular lymphoma (FL) are the most typical and dominant aggressive and indolent NHL, respectively, this study aims to develop an artificial-intelligence-enabled model to distinguish DLBCL from FL in PET/CT images as the first step to tackle this challenge. We propose a hybrid few-shot multiple-instance learning model to predict the aggressiveness of the NHL. First, rotation-based self-supervision learning (SSL) has been employed to train the encoder on a large-scale, publicly available CT image dataset. Second, hybrid instance-level features are obtained for each NHL lesion by combining deep features with the radiomics features from both PET and CT modalities. Third, instance-level features are transformed into bag-level (or patient-level) representations. Finally, bag-level representations are fed into a distance-based classifier through few-shot learning to predict NHL aggressiveness. Our model achieves an accuracy of 0.751 ± 0.008, a sensitivity of 0.787 ± 0.012, a specificity of 0.715 ± 0.013, an F1-score of 0.753 ± 0.009, and an area under the curve (AUC) of 0.795 ± 0.009 at the bag level. It outperforms the typical counterparts that use the radiomic features, random forest for feature selection, and support vector machines (SVMs) as classifiers. The three counterparts yield accuracies of 0.714 ± 0.023, 0.705 ± 0.008, and 0.698 ± 0.008, respectively. Moreover, settings of the SSL training dataset (Deep lesion) and task (rotation), hybrid CT and radiomic PET features, the pool-layer strategy of maximum, and distance-based classifier generate the best model. A hybrid few-shot multiple-instance learning model can predict lymphoma aggressiveness in PET/CT images and could be a potential tool for determining therapy strategies. Hybrid features and the combination of SSL, few-shot learning, and weakly supervised learning are the two powerful pillars of the model, and these can be expanded to other medical applications with limited samples and incomplete annotations. [ABSTRACT FROM AUTHOR]

Background: Multi-detector Computed Tomography has become an invaluable tool for the diagnosis of chronic respiratory diseases. Based on CT images, the automatic algorithm to detect the fissures and divide the lung into five lobes will help regionally quantify, amongst others, the lung density, texture, airway and, blood vessel structures, ventilation and perfusion.Methods: Sagittal adaptive fissure scanning based on the sparseness of the vessels and bronchi is employed to localize the potential fissure region. Following a Hessian matrix based line enhancement filter in the coronal slice, the shortest path is determined by means of Uniform Cost Search. Implicit surface fitting based on Radial Basis Functions is used to extract the fissure surface for lobe segmentation. By three implicit fissure surface functions, the lung is divided into five lobes. The proposed algorithm is tested by 14 datasets. The accuracy is evaluated by the mean (±S.D.), root mean square, and the maximum of the shortest Euclidian distance from the manually-defined fissure surface to that extracted by the algorithm.Results: Averaged over all datasets, the mean (±S.D.), root mean square, and the maximum of the shortest Euclidian distance are 2.05 ± 1.80, 2.46 and 7.34 mm for the right oblique fissure. The measures are 2.77 ± 2.12, 3.13 and 7.75 mm for the right horizontal fissure, 2.31 ± 1.76, 3.25 and 6.83 mm for the left oblique fissure. The fissure detection works for the data with a small lung nodule nearby the fissure and a small lung subpleural nodule. The volume and emphysema index of each lobe can be calculated. The algorithm is very fast, e.g., to finish the fissure detection and fissure extension for the dataset with 320 slices only takes around 50 seconds.Conclusions: The sagittal adaptive fissure scanning can localize the potential fissure regions quickly. After the potential region is enhanced by a Hessian based line enhancement filter, Uniform Cost Search can extract the fissures successfully in 2D. Surface fitting is able to obtain three implicit surface functions for each dataset. The current algorithm shows good accuracy, robustness and speed, may help locate the lesions into each lobe and analyze them regionally. [ABSTRACT FROM AUTHOR]

Using computational fluid dynamics (CFD) method, the feasibility of simulating transient airflow in a CT-based airway tree with more than 100 outlets for a whole respiratory period is studied, and the influence of truncations of terminal bronchi on CFD characteristics is investigated. After an airway model with 122 outlets is extracted from CT images, the transient airflow is simulated. Spatial and temporal variations of flow velocity, wall pressure, and wall shear stress are presented; the flow pattern and lobar distribution of air are gotten as well. All results are compared with those of a truncated model with 22 outlets. It is found that the flow pattern shows lobar heterogeneity that the near-wall air in the trachea is inhaled into the upper lobe while the center flow enters the other lobes, and the lobar distribution of air is significantly correlated with the outlet area ratio. The truncation decreases airflow to right and left upper lobes and increases the deviation of airflow distributions between inspiration and expiration. Simulating the transient airflow in an airway tree model with 122 bronchi using CFD is feasible. The model with more terminal bronchi decreases the difference between the lobar distributions at inspiration and at expiration. [ABSTRACT FROM AUTHOR]