Effect of 320-Row Computed Tomography Acquisition Technology on Coronary Computed Tomography Angiography-Derived Fractional Flow Reserve Based on Machine Learning: Systolic and Diastolic Scan Acquisition.

BACKGROUND: The aim of the study is to investigate the performance of coronary computed tomography angiography (CCTA)-derived fractional flow reserve (CT-FFR) in the same patient evaluated by different systolic and diastolic scans, aiming to explore whether 320-slice CT scanning acquisition protocol has an impact on CT-FFR value. METHODS: One hundred forty-six patients with suspected coronary artery stenosis who underwent CCTA examination were included into the study. The prospective electrocardiogram gated trigger sequence scan was performed and electrocardiogram editors selected 2 optimal phases of systolic phase (preset collection trigger at 25% of R-R interval) and diastolic phase (preset collection trigger at 75% of R-R interval) for reconstruction. The lowest CT-FFR value (the CT-FFR value at the distal end of each vessel) and the lesion CT-FFR value (at 2 cm distal to the stenosis) after coronary artery stenosis were calculated for each vessel. The difference of CT-FFR values between the 2 scanning techniques was compared using paired Wilcoxon signed-rank test. Pearson correlation value and Bland-Altman were performed to evaluate the consistency of CT-FFR values. RESULTS: A total of 366 coronary arteries from the remaining 122 patients were analyzed. There was no significant difference regarding the lowest CT-FFR values between systole phase and diastole phase across all vessels. In addition, there was no significant difference in the lesion CT-FFR value after coronary artery stenosis between systole phase and diastole phase across all vessels. The CT-FFR value between the 2 reconstruction techniques had excellent correlation and minimal bias in all groups. The correlation coefficient of the lesion CT-FFR values for left anterior descending branch, left circumflex branch, and right coronary artery were 0.86, 0.84, and 0.76, respectively. CONCLUSIONS: Coronary computed tomography angiography-derived fractional flow reserve based on artificial intelligence deep learning neural network has stable performance, is not affected by the acquisition phase technology of 320-slice CT scan, and has high consistency with the evaluation of hemodynamics after coronary artery stenosis.

Calcium Scoring at Coronary CT Angiography Using Deep Learning.

Background Separate noncontrast CT to quantify the coronary artery calcium (CAC) score often precedes coronary CT angiography (CTA). Quantifying CAC scores directly at CTA would eliminate the additional radiation produced at CT but remains challenging. Purpose To quantify CAC scores automatically from a single CTA scan. Materials and Methods In this retrospective study, a deep learning method to quantify CAC scores automatically from a single CTA scan was developed on training and validation sets of 292 patients and 73 patients collected from March 2019 to July 2020. Virtual noncontrast scans obtained with a spectral CT scanner were used to develop the algorithm to alleviate tedious manual annotation of calcium regions. The proposed method was validated on an independent test set of 240 CTA scans collected from three different CT scanners from August 2020 to November 2020 using the Pearson correlation coefficient, the coefficient of determination, or r2, and the Bland-Altman plot against the semiautomatic Agatston score at noncontrast CT. The cardiovascular risk categorization performance was evaluated using weighted κ based on the Agatston score (CAC score risk categories: 0-10, 11-100, 101-400, and >400). Results Two hundred forty patients (mean age, 60 years ± 11 [standard deviation]; 146 men) were evaluated. The positive correlation between the automatic deep learning CTA and semiautomatic noncontrast CT CAC score was excellent (Pearson correlation = 0.96; r2 = 0.92). The risk categorization agreement based on deep learning CTA and noncontrast CT CAC scores was excellent (weighted κ = 0.94 [95% CI: 0.91, 0.97]), with 223 of 240 scans (93%) categorized correctly. All patients who were miscategorized were in the direct neighboring risk groups. The proposed method's differences from the noncontrast CT CAC score were not statistically significant with regard to scanner (P = .15), sex (P = .051), and section thickness (P = .67). Conclusion A deep learning automatic calcium scoring method accurately quantified coronary artery calcium from CT angiography images and categorized risk. © RSNA, 2021 See also the editorial by Goldfarb and Cao et al in this issue.

Additional value of deep learning computed tomographic angiography-based fractional flow reserve in detecting coronary stenosis and predicting outcomes.

BACKGROUND: Deep learning (DL) has achieved great success in medical imaging and could be utilized for the non-invasive calculation of fractional flow reserve (FFR) from coronary computed tomographic angiography (CCTA) (CT-FFR). PURPOSE: To examine the ability of a DL-based CT-FFR in detecting hemodynamic changes of stenosis. MATERIAL AND METHODS: This study included 73 patients (85 vessels) who were suspected of coronary artery disease (CAD) and received CCTA followed by invasive FFR measurements within 90 days. The diagnostic accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and area under the receiver operating characteristics curve (AUC) were compared between CT-FFR and CCTA. Thirty-nine patients who received drug therapy instead of revascularization were followed for up to 31 months. Major adverse cardiac events (MACE), unstable angina, and rehospitalization were evaluated and compared between the study groups. RESULTS: At the patient level, CT-FFR achieved 90.4%, 93.6%, 88.1%, 85.3%, and 94.9% in accuracy, sensitivity, specificity, PPV, and NPV, respectively. At the vessel level, CT-FFR achieved 91.8%, 93.9%, 90.4%, 86.1%, and 95.9%, respectively. CT-FFR exceeded CCTA in these measurements at both levels. The vessel-level AUC for CT-FFR also outperformed that for CCTA (0.957 vs. 0.599, P < 0.0001). Patients with CT-FFR ≤0.8 had higher rates of rehospitalization (hazard ratio [HR] 4.51, 95% confidence interval [CI] 1.08-18.9) and MACE (HR 7.26, 95% CI 0.88-59.8), as well as a lower rate of unstable angina (HR 0.46, 95% CI 0.07-2.91). CONCLUSION: CT-FFR is superior to conventional CCTA in differentiating functional myocardial ischemia. In addition, it has the potential to differentiate prognoses of patients with CAD.

Using Artificial Intelligence to Detect COVID-19 and Community-acquired Pneumonia Based on Pulmonary CT: Evaluation of the Diagnostic Accuracy.

Background Coronavirus disease 2019 (COVID-19) has widely spread all over the world since the beginning of 2020. It is desirable to develop automatic and accurate detection of COVID-19 using chest CT. Purpose To develop a fully automatic framework to detect COVID-19 using chest CT and evaluate its performance. Materials and Methods In this retrospective and multicenter study, a deep learning model, the COVID-19 detection neural network (COVNet), was developed to extract visual features from volumetric chest CT scans for the detection of COVID-19. CT scans of community-acquired pneumonia (CAP) and other non-pneumonia abnormalities were included to test the robustness of the model. The datasets were collected from six hospitals between August 2016 and February 2020. Diagnostic performance was assessed with the area under the receiver operating characteristic curve, sensitivity, and specificity. Results The collected dataset consisted of 4352 chest CT scans from 3322 patients. The average patient age (±standard deviation) was 49 years ± 15, and there were slightly more men than women (1838 vs 1484, respectively; P = .29). The per-scan sensitivity and specificity for detecting COVID-19 in the independent test set was 90% (95% confidence interval [CI]: 83%, 94%; 114 of 127 scans) and 96% (95% CI: 93%, 98%; 294 of 307 scans), respectively, with an area under the receiver operating characteristic curve of 0.96 (P < .001). The per-scan sensitivity and specificity for detecting CAP in the independent test set was 87% (152 of 175 scans) and 92% (239 of 259 scans), respectively, with an area under the receiver operating characteristic curve of 0.95 (95% CI: 0.93, 0.97). Conclusion A deep learning model can accurately detect coronavirus 2019 and differentiate it from community-acquired pneumonia and other lung conditions. © RSNA, 2020 Online supplemental material is available for this article.

Evaluation of fractional flow reserve in patients with stable angina: can CT compete with angiography?

BACKGROUND: We aimed to compare the performance of FFRCT and FFRQCA in assessing the functional significance of coronary artery stenosis in patients suffering from coronary artery disease with stable angina. METHOD: A total of 101 stable coronary heart disease (CAD) patients with 181 lesions were recruited. FFRCT and FFRQCA were compared using invasive fractional flow reserve (FFR) as a reference standard. Comparisons between FFRCT and FFRQCA were conducted based on strategies of the geometric reconstruction, boundary conditions, and geometric characteristics. The performance of FFRCT and FFRQCA in detecting hemodynamic significance was also investigated. RESULTS: The performance of FFRCT and FFRQCA in discriminating hemodynamically significant lesions was compared. Good correlation and agreement with invasive FFR was found using FFRCT and FFRQCA (r = 0.809, p < 0.001 and r = 0.755, p < 0.001). A significant difference was observed in the complex coronary artery tree, in which relatively better prediction was observed using FFRCT than FFRQCA when analyzing the stenosis distributed in the middle segment of a stenotic branch (p = 0.036). Moreover, FFRCT was found to be better at predicting hemodynamically insignificant stenosis than FFRQCA (p = 0.007), while the performance of the two parameters was similar in discriminating functional significant lesions using an FFR threshold of ≤ 0.8 as a reference standard. CONCLUSION: FFRCT and FFRQCA could both accurately rule out functional insignificant lesions in stable CAD patients. FFRCT was found to be better for the noninvasive screening of CAD patients with stable angina than FFRQCA. KEY POINTS: • FFR CT and FFR QCA were both in good correlation and agreement with invasive FFR measurements. • FFR CT is superior in accuracy and consistency compared to FFR QCA in patients with stenoses distributed in left coronary artery. • The noninvasive nature of FFR CT could provide potential benefit for stable CAD patients on disease management.

Precise diagnosis of intracranial hemorrhage and subtypes using a three-dimensional joint convolutional and recurrent neural network.

OBJECTIVES: To evaluate the performance of a novel three-dimensional (3D) joint convolutional and recurrent neural network (CNN-RNN) for the detection of intracranial hemorrhage (ICH) and its five subtypes (cerebral parenchymal, intraventricular, subdural, epidural, and subarachnoid) in non-contrast head CT. METHODS: A total of 2836 subjects (ICH/normal, 1836/1000) from three institutions were included in this ethically approved retrospective study, with a total of 76,621 slices from non-contrast head CT scans. ICH and its five subtypes were annotated by three independent experienced radiologists, with majority voting as reference standard for both the subject level and the slice level. Ninety percent of data was used for training and validation, and the rest 10% for final evaluation. A joint CNN-RNN classification framework was proposed, with the flexibility to train when subject-level or slice-level labels are available. The predictions were compared with the interpretations from three junior radiology trainees and an additional senior radiologist. RESULTS: It took our algorithm less than 30 s on average to process a 3D CT scan. For the two-type classification task (predicting bleeding or not), our algorithm achieved excellent values (≥ 0.98) across all reporting metrics on the subject level. For the five-type classification task (predicting five subtypes), our algorithm achieved > 0.8 AUC across all subtypes. The performance of our algorithm was generally superior to the average performance of the junior radiology trainees for both two-type and five-type classification tasks. CONCLUSIONS: The proposed method was able to accurately detect ICH and its subtypes with fast speed, suggesting its potential for assisting radiologists and physicians in their clinical diagnosis workflow. KEY POINTS: • A 3D joint CNN-RNN deep learning framework was developed for ICH detection and subtype classification, which has the flexibility to train with either subject-level labels or slice-level labels. • This deep learning framework is fast and accurate at detecting ICH and its subtypes. • The performance of the automated algorithm was superior to the average performance of three junior radiology trainees in this work, suggesting its potential to reduce initial misinterpretations.

Hemodynamics analysis of the serial stenotic coronary arteries.

Coronary arterial stenoses, particularly serial stenoses in a single branch, are responsible for complex hemodynamic properties of the coronary arterial trees, and the uncertain prognosis of invasive intervention. Critical information of the blood flow redistribution in the stenotic arterial segments is required for the adequate treatment planning. Therefore, in this study, an image based non-invasive functional assessment is performed to investigate the hemodynamic significances of serial stenoses. Twenty patient-specific coronary arterial trees with different combinations of stenoses were reconstructed from the computer tomography angiography for the evaluation of the hemodynamics. Our results showed that the computed FFR based on CTA images (FFRCT) pullback curves with wall shear stress (WSS) distribution could provide more effectively examine the physiological significance of the locations of the segmental narrowing and the curvature of the coronary arterial segments. The paper thus provides the diagnostic efficacy of FFRCT pullback curve for noninvasive quantification of the hemodynamics of stenotic coronary arteries with serial lesions, compared to the gold standard invasive FFR, to provide a reliable physiological assessment of significant amount of coronary artery stenosis. Further, we were also able to demonstrate the potential of carrying out virtual revascularization, to enable more precise PCI procedures and improve their outcomes.

A Novel Method of Estimating Small Airway Disease Using Inspiratory-to-Expiratory Computed Tomography.

BACKGROUND: Disease accumulates in the small airways without being detected by conventional measurements. OBJECTIVES: To quantify small airway disease using a novel computed tomography (CT) inspiratory-to-expiratory approach called the disease probability measure (DPM) and to investigate the association with pulmonary function measurements. METHODS: Participants from the population-based CanCOLD study were evaluated using full-inspiration/full-expiration CT and pulmonary function measurements. Full-inspiration and full-expiration CT images were registered, and each voxel was classified as emphysema, gas trapping (GasTrap) related to functional small airway disease, or normal using two classification approaches: parametric response map (PRM) and DPM (VIDA Diagnostics, Inc., Coralville, IA, USA). RESULTS: The participants included never-smokers (n = 135), at risk (n = 97), Global Initiative for Chronic Obstructive Lung Disease I (GOLD I) (n = 140), and GOLD II chronic obstructive pulmonary disease (n = 96). PRMGasTrap and DPMGasTrap measurements were significantly elevated in GOLD II compared to never-smokers (p < 0.01) and at risk (p < 0.01), and for GOLD I compared to at risk (p < 0.05). Gas trapping measurements were significantly elevated in GOLD II compared to GOLD I (p < 0.0001) using the DPM classification only. Overall, DPM classified significantly more voxels as gas trapping than PRM (p < 0.0001); a spatial comparison revealed that the expiratory CT Hounsfield units (HU) for voxels classified as DPMGasTrap but PRMNormal (PRMNormal- DPMGasTrap = -785 ± 72 HU) were significantly reduced compared to voxels classified normal by both approaches (PRMNormal-DPMNormal = -722 ± 89 HU; p < 0.0001). DPM and PRMGasTrap measurements showed similar, significantly associations with forced expiratory volume in 1 s (FEV1) (p < 0.01), FEV1/forced vital capacity (p < 0.0001), residual volume/total lung capacity (p < 0.0001), bronchodilator response (p < 0.0001), and dyspnea (p < 0.05). CONCLUSION: CT inspiratory-to-expiratory gas trapping measurements are significantly associated with pulmonary function and symptoms. There are quantitative and spatial differences between PRM and DPM classification that need pathological investigation.

Computed tomography predictors of response to endobronchial valve lung reduction treatment. Comparison with Chartis.

RATIONALE: Chartis Pulmonary Assessment System (Pulmonx Inc., Redwood, CA) is an invasive procedure used to assess collateral ventilation and select candidates for valve-based lung volume reduction (LVR) therapy. Quantitative computed tomography (QCT) is a potential alternative to Chartis and today consists primarily of assessing fissure integrity (FI). OBJECTIVES: In this retrospective analysis, we aimed to determine QCT predictors of LVR outcome and compare the QCT model with Chartis in selecting likely responders to valve-based LVR treatment. METHODS: Baseline CT scans of 146 subjects with severe emphysema who underwent endobronchial valve LVR were analyzed retrospectively using dedicated lung quantitative imaging software (Apollo; VIDA Diagnostics, Coralville, IA). A lobar volume reduction greater than 350 ml at 3 months was considered to be indicative of positive response to treatment. Thirty-four CT baseline variables, including quantitative measurements of FI, density, and vessel volumetry, were used to feed a multiple logistic regression analysis to find significant predictors of LVR outcome. The primary predictors were then used in 33 datasets with Chartis results to evaluate the relative performance of QCT versus Chartis. MEASUREMENTS AND MAIN RESULTS: FI (P < 0.0001) and low attenuation clusters (P = 0.01) measured in the treated lobe and vascular volumetric percentage of patient's detected smallest vessels (P = 0.02) were identified as the primary QCT predictors of LVR outcome. Accuracy for QCT patient selection based on these primary predictors was comparable to Chartis (78.8 vs. 75.8%). CONCLUSIONS: Quantitative CT led to comparable results to Chartis for classifying LVR and is a promising tool to effectively select patients for valve-based LVR procedures.

Pulmonary Microvascular Blood Flow in Mild Chronic Obstructive Pulmonary Disease and Emphysema. The MESA COPD Study.

RATIONALE: Smoking-related microvascular loss causes end-organ damage in the kidneys, heart, and brain. Basic research suggests a similar process in the lungs, but no large studies have assessed pulmonary microvascular blood flow (PMBF) in early chronic lung disease. OBJECTIVES: To investigate whether PMBF is reduced in mild as well as more severe chronic obstructive pulmonary disease (COPD) and emphysema. METHODS: PMBF was measured using gadolinium-enhanced magnetic resonance imaging (MRI) among smokers with COPD and control subjects age 50 to 79 years without clinical cardiovascular disease. COPD severity was defined by standard criteria. Emphysema on computed tomography (CT) was defined by the percentage of lung regions below -950 Hounsfield units (-950 HU) and by radiologists using a standard protocol. We adjusted for potential confounders, including smoking, oxygenation, and left ventricular cardiac output. MEASUREMENTS AND MAIN RESULTS: Among 144 participants, PMBF was reduced by 30% in mild COPD, by 29% in moderate COPD, and by 52% in severe COPD (all P < 0.01 vs. control subjects). PMBF was reduced with greater percentage emphysema-950HU and radiologist-defined emphysema, particularly panlobular and centrilobular emphysema (all P ≤ 0.01). Registration of MRI and CT images revealed that PMBF was reduced in mild COPD in both nonemphysematous and emphysematous lung regions. Associations for PMBF were independent of measures of small airways disease on CT and gas trapping largely because emphysema and small airways disease occurred in different smokers. CONCLUSIONS: PMBF was reduced in mild COPD, including in regions of lung without frank emphysema, and may represent a distinct pathological process from small airways disease. PMBF may provide an imaging biomarker for therapeutic strategies targeting the pulmonary microvasculature.

Patterns of Emphysema Heterogeneity.

BACKGROUND: Although lobar patterns of emphysema heterogeneity are indicative of optimal target sites for lung volume reduction (LVR) strategies, the presence of segmental, or sublobar, heterogeneity is often underappreciated. OBJECTIVE: The aim of this study was to understand lobar and segmental patterns of emphysema heterogeneity, which may more precisely indicate optimal target sites for LVR procedures. METHODS: Patterns of emphysema heterogeneity were evaluated in a representative cohort of 150 severe (GOLD stage III/IV) chronic obstructive pulmonary disease (COPD) patients from the COPDGene study. High-resolution computerized tomography analysis software was used to measure tissue destruction throughout the lungs to compute heterogeneity (≥15% difference in tissue destruction) between (inter-) and within (intra-) lobes for each patient. Emphysema tissue destruction was characterized segmentally to define patterns of heterogeneity. RESULTS: Segmental tissue destruction revealed interlobar heterogeneity in the left lung (57%) and right lung (52%). Intralobar heterogeneity was observed in at least one lobe of all patients. No patient presented true homogeneity at a segmental level. There was true homogeneity across both lungs in 3% of the cohort when defining heterogeneity as ≥30% difference in tissue destruction. CONCLUSION: Many LVR technologies for treatment of emphysema have focused on interlobar heterogeneity and target an entire lobe per procedure. Our observations suggest that a high proportion of patients with emphysema are affected by interlobar as well as intralobar heterogeneity. These findings prompt the need for a segmental approach to LVR in the majority of patients to treat only the most diseased segments and preserve healthier ones.

One-stop patient-specific myocardial blood flow quantification technique based on allometric scaling law.

Establishing a patient-specific and non-invasive technique to derive blood flow as well as coronary structural information from one single cardiac CT imaging modality. 336 patients with chest pain or ST segment depression on electrocardiogram were retrospectively enrolled. All patients underwent adenosine-stressed dynamic CT myocardial perfusion imaging (CT-MPI) and coronary computed tomography angiography (CCTA) in sequence. Relationship between myocardial mass (M) and blood flow (Q), defined as log(Q) = b · log(M) + log(Q0), was explored based on the general allometric scaling law. We used 267 patients to obtain the regression results and found strong linear relationship between M (gram) and Q (mL/min) (b = 0.786, log(Q0) = 0.546, r = 0.704; p < 0.001). We Also found this correlation was applicable for patients with either normal or abnormal myocardial perfusion (p < 0.001). Datasets from the other 69 patients were used to validate this M-Q correlation and found the patient-specific blood flow could be accurately estimated from CCTA compared to that measured from CT-MPI (146.480 ± 39.607 vs 137.967 ± 36.227, r = 0.816, and 146.480 ± 39.607 vs 137.967 ± 36.227, r = 0.817, for the left ventricle region and LAD-subtended region, respectively, all unit in mL/min). In conclusion, we established a technique to provide general and patient-specific myocardial mass-blood flow correlation obeyed to allometric scaling law. Blood flow information could be directly derived from structural information acquired from CCTA.

Coronary Computed Tomography Angiography for the Assessment of Sirolimus-Eluting Resorbable Magnesium Scaffold.

BACKGROUND: Little evidence to date has described the feasibility and diagnostic accuracy of coronary computed tomography angiography (CCTA) with noninvasive fractional flow reserve (CT-FFR) in coronary vessels with resorbable magnesium scaffold (RMS). METHODS: The SHERPA-MAGIC is a prospective study enrolling patients receiving RMS. The present analysis considered patients undergoing CCTA 18 months after the index procedure. CCTA images were employed to investigate reabsorption status, luminal measurements, and noninvasive FFR. Three-year follow-up was available for all patients. RESULTS: Overall, 26 patients with a total of 29 coronary arteries treated with 35 RMS were considered. The most frequently involved vessel was left anterior descendent (LAD). Median stent length was 25 (20-25) mm, with a median diameter of 3 (3-3.5) mm. At 18-month CCTA, all scaffolded segments were patent. Complete RMS reabsorption was observed in 27 (93%, 95% CI 77-99%) cases. Median minimal lumen diameter (MLD) and area (MLA) of the scaffolded segments were 2.5 [2.1-2.8] mm and 6.4 [4.4-8.4] mm2, respectively. Median CT-FFR was 0.88 [0.81-0.91]. Only one (3.5%) vessel showed a flow-limiting CT-FFR value ≤0.80. During the 3-year follow-up, only one (4%) adverse event was observed. Conclusions: In patients undergoing RMS implantation, CCTA including noninvasive CT-FFR evaluation is feasible and allows investigation of long-term RMS performance.

One deep learning local-global model based on CT imaging to differentiate between nodular cryptococcosis and lung cancer which are hard to be diagnosed.

OBJECTIVES: We aim to evaluate a deep learning (DL) model and radiomic model for preoperative differentiation of nodular cryptococcosis from solitary lung cancer in patients with malignant features on CT images. MATERIALS AND METHODS: We retrospectively recruited 319 patients with solitary pulmonary nodules and suspicious signs of malignancy from three hospitals. All lung nodules were resected, and one by one radiologic-pathologic correlation was performed. A three-dimensional DL model was used for tumor segmentation and extraction of three-dimensional radiomic features. We used the Max-Relevance and Min-Redundancy algorithm and the eXtreme Gradient Boosting algorithm to select the nodular radiomics features. We proposed a DL local-global model, a DL local model and radiomic model to preoperatively differentiate nodular cryptococcosis from solitary lung cancer. The DL local-global model includes information of both nodules and the whole lung, while the DL local model only includes information of solitary lung nodules. Five-fold cross-validation was used to select and validate these models. The prediction performance of the model was evaluated using receiver operating characteristic curve (ROC) and calibration curve. A new loss function was applied in our deep learning framework to optimize the area under the ROC curve (AUC) directly. RESULTS: 295 patients were enrolled and they were non-symptomatic, with negative tumor markers and fungus markers in blood tests. These patients have not been diagnosed by the combination of CT imaging, laboratory results and clinical data. The lung volume was slightly larger in patients with lung cancers than that in patients with cryptococcosis (3552.8 ± 1184.6 ml vs 3491.9 ± 1017.8 ml). The DL local-global model achieved the best performance in differentiating between nodular cryptococcosis and lung cancer (area under the curve [AUC] = 0.88), which was higher than that of the DL local model (AUC = 0.84) and radiomic (AUC = 0.79) model. CONCLUSION: The DL local-global model is a non-invasive diagnostic tool to differentiate between nodular cryptococcosis and lung cancer nodules which are hard to be diagnosed by the combination of CT imaging, laboratory results and clinical data, and overtreatment may be avoided.

End-to-end multimodal image registration via reinforcement learning.

Multimodal image registration is a vital initial step in several medical image applications for providing complementary information from different data modalities. Since images with different modalities do not exhibit the same characteristics, finding their accurate correspondences remains a challenge. For convolutional multimodal registration methods, two components are quite significant: descriptive image feature as well as the suited similarity metric. However, these two components are often custom-designed and are infeasible to the high diversity of tissue appearance across modalities. In this paper, we translate image registration into a decision-making problem, where registration is achieved via an artificial agent trained by asynchronous reinforcement learning. More specifically, convolutional long-short-term-memory is incorporated after stacked convolutional layers in this method to extract spatial-temporal image features and learn the similarity metric implicitly. A customized reward function driven by landmark error is advocated to guide the agent to the correct registration direction. A Monte Carlo rollout strategy is also leveraged to perform as a look-ahead inference in the testing stage, to increase registration accuracy further. Experiments on paired CT and MR images of patients diagnosed as nasopharyngeal carcinoma demonstrate that our method achieves state-of-the-art performance in medical image registration.

Learning physical properties in complex visual scenes: An intelligent machine for perceiving blood flow dynamics from static CT angiography imaging.

Humans perceive physical properties such as motion and elastic force by observing objects in visual scenes. Recent research has proven that computers are capable of inferring physical properties from camera images like humans. However, few studies perceive the physical properties in more complex environment, i.e. humans have difficulty estimating physical quantities directly from the visual observation, or encounter difficulty visualizing the physical process in mind according to their daily experiences. As an appropriate example, fractional flow reserve (FFR), which measures the blood pressure difference across the vessel stenosis, becomes an important physical quantitative value determining the likelihood of myocardial ischemia in clinical coronary intervention procedure. In this study, we propose a novel deep neural network solution (TreeVes-Net) that allows machines to perceive FFR values directly from static coronary CT angiography images. Our framework fully utilizes a tree-structured recurrent neural network (RNN) with a coronary representation encoder. The encoder captures coronary geometric information providing the blood fluid-related representation. The tree-structured RNN builds a long-distance spatial dependency of blood flow information inside the coronary tree. The experiments performed on 13000 synthetic coronary trees and 180 real coronary trees from clinical patients show that the values of the area under ROC curve (AUC) are 0.92 and 0.93 under two clinical criterions. These results can demonstrate the effectiveness of our framework and its superiority to seven FFR computation methods based on machine learning.

Learning tree-structured representation for 3D coronary artery segmentation.

Extensive research has been devoted to the segmentation of the coronary artery. However, owing to its complex anatomical structure, it is extremely challenging to automatically segment the coronary artery from 3D coronary computed tomography angiography (CCTA). Inspired by recent ideas to use tree-structured long short-term memory (LSTM) to model the underlying tree structures for NLP tasks, we propose a novel tree-structured convolutional gated recurrent unit (ConvGRU) model to learn the anatomical structure of the coronary artery. However, unlike tree-structured LSTM proposed for semantic relatedness as well as sentiment classification in natural language processing, our tree-structured ConvGRU model considers the local spatial correlations in the input data as the convolutions are used for input-to-state as well as state-to-state transitions, thus more suitable for image analysis. To conduct voxel-wise segmentation, a tree-structured segmentation framework is presented. It consists of a fully convolutional network (FCN) for multi-scale discriminative feature extraction and the final prediction, and a tree-structured ConvGRU layer for anatomical structure modeling. The proposed framework is extensively evaluated on four large-scale 3D CCTA dataset (the largest to the best of our knowledge), and experiments show that our method is more accurate as well as efficient, compared with other coronary artery segmentation approaches.

Mass preserving nonrigid registration of CT lung images using cubic B-spline.

The authors propose a nonrigid image registration approach to align two computed-tomography (CT)-derived lung datasets acquired during breath-holds at two inspiratory levels when the image distortion between the two volumes is large. The goal is to derive a three-dimensional warping function that can be used in association with computational fluid dynamics studies. In contrast to the sum of squared intensity difference (SSD), a new similarity criterion, the sum of squared tissue volume difference (SSTVD), is introduced to take into account changes in reconstructed Hounsfield units (scaled attenuation coefficient, HU) with inflation. This new criterion aims to minimize the local tissue volume difference within the lungs between matched regions, thus preserving the tissue mass of the lungs if the tissue density is assumed to be relatively constant. The local tissue volume difference is contributed by two factors: Change in the regional volume due to the deformation and change in the fractional tissue content in a region due to inflation. The change in the regional volume is calculated from the Jacobian value derived from the warping function and the change in the fractional tissue content is estimated from reconstructed HU based on quantitative CT measures. A composite of multilevel B-spline is adopted to deform images and a sufficient condition is imposed to ensure a one-to-one mapping even for a registration pair with large volume difference. Parameters of the transformation model are optimized by a limited-memory quasi-Newton minimization approach in a multiresolution framework. To evaluate the effectiveness of the new similarity measure, the authors performed registrations for six lung volume pairs. Over 100 annotated landmarks located at vessel bifurcations were generated using a semiautomatic system. The results show that the SSTVD method yields smaller average landmark errors than the SSD method across all six registration pairs.

An iterative multi-path fully convolutional neural network for automatic cardiac segmentation in cine MR images.

PURPOSE: Segmentation of the left ventricle (LV), right ventricle (RV) cavities and the myocardium (MYO) from cine cardiac magnetic resonance (MR) images is an important step for diagnosis and monitoring cardiac diseases. Spatial context information may be highly beneficial for segmentation performance improvement. To this end, this paper proposes an iterative multi-path fully convolutional network (IMFCN) to effectively leverage spatial context for automatic cardiac segmentation in cine MR images. METHODS: To effectively leverage spatial context information, the proposed IMFCN explicitly models the interslice spatial correlations using a multi-path late fusion strategy. First, the contextual inputs including both the adjacent slices and the already predicted mask of the above adjacent slice are processed by independent feature-extraction paths. Then, an atrous spatial pyramid pooling (ASPP) module is employed at the feature fusion process to combine the extracted high-level contextual features in a more effective way. Finally, deep supervision (DS) and batch-wise class re-weighting mechanism are utilized to enhance the training of the proposed network. RESULTS: The proposed IMFCN was evaluated and analyzed on the MICCAI 2017 automatic cardiac diagnosis challenge (ACDC) dataset. On the held-out training dataset reserved for testing, our method effectively improved its counterparts that without spatial context and that with spatial context but using an early fusion strategy. On the 50 subjects test dataset, our method achieved Dice similarity coefficient of 0.935, 0.920, and 0.905, and Hausdorff distance of 7.66, 12.10, and 8.80 mm for LV, RV, and MYO, respectively, which are comparable or even better than the state-of-the-art methods of ACDC Challenge. In addition, to explore the applicability to other datasets, the proposed IMFCN was retrained on the Sunnybrook dataset for LV segmentation and also produced comparable performance to the state-of-the-art methods. CONCLUSIONS: We have presented an automatic end-to-end fully convolutional architecture for accurate cardiac segmentation. The proposed method provides an effective way to leverage spatial context in a two-dimensional manner and results in precise and consistent segmentation results.

Automated anatomical labeling of coronary arteries via bidirectional tree LSTMs.

PURPOSE: Automated anatomical labeling facilitates the diagnostic process for physicians and radiologists. One of the challenges in automated anatomical labeling problems is the robustness to handle the large individual variability inherited in human anatomy. A novel deep neural network framework, referred to Tree Labeling Network (TreeLab-Net), is proposed to resolve this problem in this work. METHODS: A multi-layer perceptron (MLP) encoder network and a bidirectional tree-structural long short-term memory (Bi-TreeLSTM) are combined to construct the TreeLab-Net. Vessel spatial locations and directions are selected as features, where a spherical coordinate transform is utilized to normalize vessel spatial variations. The dataset includes 436 coronary computed tomography angiography images. Tenfold cross-validation is performed for evaluation. RESULTS: The precision-recall curve of TreeLab-Net shows that the four main branch classes, LM, LAD, LCX and RCA, have the area under the curve (AUC) higher than 97%. Other major side branch classes, D, OM, and R-PLB, also have AUC higher than 90%. Comparing with four other methods (i.e., AdaBoost, MLP, Up-to-Down and Down-to-Up TreeLSTM), the TreeLab-Net achieves higher F1 scores with less topological errors. CONCLUSION: The TreeLab-Net is able to capture the characteristics of tree structures by learning the spatial and topological dependencies of blood vessels effectively. The results demonstrate that TreeLab-Net is able to yield competitive performances on a large dataset with great variance among subjects.