Anatomical pulmonary magnetic resonance imaging segmentation for regional structure-function measurements of asthma

Specific Ventilation in Severe Asthma Evaluated with Noncontrast Tidal Breathing 1H MRI

Purpose To determine if proton (1H) MRI-derived specific ventilation is responsive to bronchodilator (BD) therapy and associated with clinical biomarkers of type 2 airway inflammation and airways dysfunction in severe asthma. Materials and Methods In this prospective study, 27 participants with severe asthma (mean age, 52 years ± 9 [SD]; 17 female, 10 male) and seven healthy controls (mean age, 47 years ± 16; five female, two male), recruited between 2018 and 2021, underwent same-day spirometry, respiratory oscillometry, and tidal breathing 1H MRI. Participants with severe asthma underwent all assessments before and after BD therapy, and type 2 airway inflammatory biomarkers were determined (blood eosinophil count, sputum eosinophil percentage, sputum eosinophil-free granules, and fraction of exhaled nitric oxide) to generate a cumulative type 2 biomarker score. Specific ventilation was derived from tidal breathing 1H MRI and its response to BD therapy, and relationships with biomarkers of type 2 airway inflammation and airway dysfunction were evaluated. Results Mean MRI specific ventilation improved with BD inhalation (from 0.07 ± 0.04 to 0.11 ± 0.04, P < .001). Post-BD MRI specific ventilation (P = .046) and post-BD change in MRI specific ventilation (P = .006) were greater in participants with asthma with type 2 low biomarkers compared with participants with type 2 high biomarkers of airway inflammation. Post-BD change in MRI specific ventilation was correlated with change in forced expiratory volume in 1 second (r = 0.40, P = .04), resistance at 5 Hz (r = -0.50, P = .01), resistance at 19 Hz (r = -0.42, P = .01), reactance area (r = -0.54, P < .01), and reactance at 5 Hz (r = 0.48, P = .01). Conclusion Specific ventilation evaluated with tidal breathing 1H MRI was responsive to BD therapy and was associated with clinical biomarkers of airways disease in participants with severe asthma. Keywords: MRI, Severe Asthma, Ventilation, Type 2 Inflammation Supplemental material is available for this article. © RSNA, 2023 See also the commentary by Moore and Chandarana in this issue.

Cardiac MRI segmentation with sparse annotations: Ensembling deep learning uncertainty and shape priors

The performance of deep learning for cardiac magnetic resonance imaging (MRI) segmentation is oftentimes degraded when using small datasets and sparse annotations for training or adapting a pre-trained model to previously unseen datasets. Here, we developed and evaluated an approach to addressing some of these issues to facilitate broader use of deep learning for short-axis cardiac MRI segmentation. We developed a globally optimal label fusion (GOLF) algorithm that enforced spatial smoothness to generate consensus segmentation from segmentation predictions provided by a deep learning ensemble algorithm. The GOLF consensus was entered into an uncertainty-guided coupled continuous kernel cut (ugCCKC) algorithm that employed normalized cut, image-grid continuous regularization, and "nesting" and circular shape priors of the left ventricular myocardium (LVM) and cavity (LVC). In addition, the uncertainty measurements derived from the segmentation predictions were used to constrain the similarity of GOLF and final segmentation. We optimized ugCCKC through upper bound relaxation, for which we developed an efficient coupled continuous max-flow algorithm implemented in an iterative manner. We showed that GOLF yielded average symmetric surface distance (ASSD) 0.2-0.8 mm lower than an averaging method with higher or similar Dice similarity coefficient (DSC). We also demonstrated that ugCCKC incorporating the shape priors improved DSC by 0.01-0.05 and reduced ASSD by 0.1-0.9 mm. In addition, we integrated GOLF and ugCCKC into a deep learning ensemble algorithm by refining the segmentation of an unannotated dataset and using the refined segmentation to update the trained models. With the proposed framework, we demonstrated the capability of using relatively small datasets (5-10 subjects) with sparse (5-25% slices labeled) annotations to train a deep learning algorithm, while achieving DSC of 0.871-0.893 for LVM and 0.933-0.959 for LVC on the LVQuan dataset, and these were 0.844-0.871 for LVM and 0.923-0.931 for LVC on the ACDC dataset. Furthermore, we showed that the proposed approach can be adapted to substantially alleviate the domain shift issue. Moreover, we calculated a number of commonly used LV function measurements using the derived segmentation and observed strong correlations (Pearson r=0.77-1.00, p<0.001) between algorithm and manual LV function analyses. These results suggest that the developed approaches can be used to facilitate broader application of deep learning in research and clinical cardiac MR imaging workflow.

Ultra-short echo-time magnetic resonance imaging lung segmentation with under-Annotations and domain shift

Ultra-short echo-time (UTE) magnetic resonance imaging (MRI) provides enhanced visualization of pulmonary structural and functional abnormalities and has shown promise in phenotyping lung disease. Here, we describe the development and evaluation of a lung segmentation approach to facilitate UTE MRI methods for patient-based imaging. The proposed approach employs a k-means algorithm in kernel space for pair-wise feature clustering and imposes image domain continuous regularization, coined as continuous kernel k-means (CKKM). The high-order CKKM algorithm was simplified through upper bound relaxation and solved within an iterative continuous max-flow framework. We combined the CKKM with U-net and atlas-based approaches and comprehensively evaluated the performance on 100 images from 25 patients with asthma and bronchial pulmonary dysplasia enrolled at Robarts Research Institute (Western University, London, Canada) and Centre Hospitalier Universitaire (Sainte-Justine, Montreal, Canada). For U-net, we trained the network five times on a mixture of five different images with under-annotations and applied the model to 64 images from the two centres. We also trained a U-net on five images with full and brush annotations from one centre, and tested the model on 32 images from the other centre. For an atlas-based approach, we employed three atlas images to segment 64 target images from the two centres through straightforward atlas registration and label fusion. We applied the CKKM algorithm to the baseline U-net and atlas outputs and refined the initial segmentation through multi-volume image fusion. The integration of CKKM substantially improved baseline results and yielded, with minimal computational cost, segmentation accuracy, and precision that were greater than some state-of-the-art deep learning models and similar to experienced observer manual segmentation. This suggests that deep learning and atlas-based approaches may be utilized to segment UTE MRI datasets using relatively small training datasets with under-annotations.

Pulmonary Ventilation Maps Generated with Free-breathing Proton MRI and a Deep Convolutional Neural Network

Background Hyperpolarized noble gas MRI helps measure lung ventilation, but clinical translation remains limited. Free-breathing proton MRI may help quantify lung function using existing MRI systems without contrast material and may assist in providing information about ventilation not visible to the eye or easily extracted with segmentation methods. Purpose To explore the use of deep convolutional neural networks (DCNNs) to generate synthetic MRI ventilation scans from free-breathing MRI (deep learning [DL] ventilation MRI)-derived specific ventilation maps as a surrogate of noble gas MRI and to validate this approach across a wide range of lung diseases. Materials and Methods In this secondary analysis of prospective trials, 114 paired noble gas MRI and two-dimensional free-breathing MRI scans were obtained in healthy volunteers with no history of chronic or acute respiratory disease and in study participants with a range of different obstructive lung diseases, including asthma, bronchiectasis, chronic obstructive pulmonary disease, and non-small-cell lung cancer between September 2013 and April 2018 (ClinicalTrials.gov identifiers: NCT03169673, NCT02351141, NCT02263794, NCT02282202, NCT02279329, and NCT02002052). A U-Net-based DCNN model was trained to map free-breathing proton MRI to hyperpolarized helium 3 (³He) MRI ventilation and validated using a sixfold validation. During training, the DCNN ventilation maps were compared with noble gas MRI scans using the Pearson correlation coefficient (r) and mean absolute error. DCNN ventilation images were segmented for ventilation and ventilation defects and were compared with noble gas MRI scans using the Dice similarity coefficient (DSC). Relationships were evaluated with the Spearman correlation coefficient (r_S). Results One hundred fourteen study participants (mean age, 56 years ± 15 [standard deviation]; 66 women) were evaluated. As compared with ³He MRI, DCNN model ventilation maps had a mean r value of 0.87 ± 0.08. The mean DSC for DL ventilation MRI and ³He MRI ventilation was 0.91 ± 0.07. The ventilation defect percentage for DL ventilation MRI was highly correlated with ³He MRI ventilation defect percentage (r_S = 0.83, P < .001, mean bias = -2.0% ± 5). Both DL ventilation MRI (r_S = -0.51, P < .001) and ³He MRI (r_S = -0.61, P < .001) ventilation defect percentage were correlated with the forced expiratory volume in 1 second. The DCNN model required approximately 2 hours for training and approximately 1 second to generate a ventilation map. Conclusion In participants with diverse pulmonary pathologic findings, deep convolutional neural networks generated ventilation maps from free-breathing proton MRI trained with a hyperpolarized noble-gas MRI ventilation map data set. The maps showed correlation with noble gas MRI ventilation and pulmonary function measurements. © RSNA, 2020 See also the editorial by Vogel-Claussen in this issue.

A Voxel-Based Fully Convolution Network and Continuous Max-Flow for Carotid Vessel-Wall-Volume Segmentation From 3D Ultrasound Images

Vessel-wall-volume (VWV) is an important three-dimensional ultrasound (3DUS) metric used in the assessment of carotid plaque burden and monitoring changes in carotid atherosclerosis in response to medical treatment. To generate the VWV measurement, we proposed an approach that combined a voxel-based fully convolution network (Voxel-FCN) and a continuous max-flow module to automatically segment the carotid media-adventitia (MAB) and lumen-intima boundaries (LIB) from 3DUS images. Voxel-FCN includes an encoder consisting of a general 3D CNN and a 3D pyramid pooling module to extract spatial and contextual information, and a decoder using a concatenating module with an attention mechanism to fuse multi-level features extracted by the encoder. A continuous max-flow algorithm is used to improve the coarse segmentation provided by the Voxel-FCN. Using 1007 3DUS images, our approach yielded a Dice-similarity-coefficient (DSC) of 93.2±3.0% for the MAB in the common carotid artery (CCA), and 91.9±5.0% in the bifurcation by comparing algorithm and expert manual segmentations. We achieved a DSC of 89.5±6.7% and 89.3±6.8% for the LIB in the CCA and the bifurcation respectively. The mean errors between the algorithm-and manually-generated VWVs were 0.2±51.2 mm³ for the CCA and -4.0±98.2 mm³ for the bifurcation. The algorithm segmentation accuracy was comparable to intra-observer manual segmentation but our approach required less than 1s, which will not alter the clinical work-flow as 10s is required to image one side of the neck. Therefore, we believe that the proposed method could be used clinically for generating VWV to monitor progression and regression of carotid plaques.

A framework for Fourier-decomposition free-breathing pulmonary 1 H MRI ventilation measurements

PURPOSE

To develop a rapid Fourier decomposition (FD) free-breathing pulmonary ¹ H MRI (FDMRI) image processing and biomarker pipeline for research use.

METHODS

We acquired MRI in 20 asthmatic subjects using a balanced steady-state free precession (bSSFP) sequence optimized for ventilation imaging. 2D ¹ H MRI series were segmented by enforcing the spatial similarity between adjacent images and the right-to-left lung volume-ratio. The segmented lung series were co-registered using a coarse-to-fine deformable registration framework that used dual optimization techniques. All pairwise registrations were implemented in parallel and FD was performed to generate 2D ventilation-weighted maps and ventilation-defect-percent (VDP). Lung segmentation and registration accuracy were evaluated by comparing algorithm and manual lung-masks, deformed manual lung-masks, and fiducials in the moving and fixed images using Dice-similarity-coefficient (DSC), mean-absolute-distance (MAD), and target-registration-error (TRE). The relationship of FD-VDP and ³ He-VDP was evaluated using the Pearson-correlation-coefficient (r) and Bland Altman analysis. Algorithm reproducibility was evaluated using the coefficient-of-variation (CoV) and intra-class-correlation-coefficient (ICC) for segmentation, registration, and FD-VDP components.

RESULTS

For lung segmentation, there was a DSC of 95 ± 1.5% and MAD of 2.3 ± 0.5 mm, and for registration there was a DSC of 97 ± 0.8%, MAD of 1.6 ± 0.4 mm and TRE of 3.6 ± 1.2 mm. Reproducibility for segmentation DSC (CoV/ICC = 0.5%/0.92), registration TRE (CoV/ICC = 0.4%/0.98), and FD-VDP (Cov/ICC = 3.9%/0.97) was high. The pipeline required 10 min/subject. FD-VDP was correlated with ³ He-VDP (r = 0.69, P < 0.001) although there was a bias toward lower FD-VDP (bias = -4.9%).

CONCLUSIONS

We developed and evaluated a pipeline that provides a rapid and precise method for FDMRI ventilation maps.

Development of a pulmonary imaging biomarker pipeline for phenotyping of chronic lung disease

We designed and generated pulmonary imaging biomarker pipelines to facilitate high-throughput research and point-of-care use in patients with chronic lung disease. Image processing modules and algorithm pipelines were embedded within a graphical user interface (based on the .NET framework) for pulmonary magnetic resonance imaging (MRI) and x-ray computed-tomography (CT) datasets. The software pipelines were generated using C++ and included: (1) inhaled He3/Xe129 MRI ventilation and apparent diffusion coefficients, (2) CT-MRI coregistration for lobar and segmental ventilation and perfusion measurements, (3) ultrashort echo-time H1 MRI proton density measurements, (4) free-breathing Fourier-decomposition H1 MRI ventilation/perfusion and free-breathing H1 MRI specific ventilation, (5) multivolume CT and MRI parametric response maps, and (6) MRI and CT texture analysis and radiomics. The image analysis framework was implemented on a desktop workstation/tablet to generate biomarkers of regional lung structure and function related to ventilation, perfusion, lung tissue texture, and integrity as well as multiparametric measures of gas trapping and airspace enlargement. All biomarkers were generated within 10 min with measurement reproducibility consistent with clinical and research requirements. The resultant pulmonary imaging biomarker pipeline provides real-time and automated lung imaging measurements for point-of-care and high-throughput research.

Free-breathing Pulmonary MR Imaging to Quantify Regional Ventilation

Purpose To measure regional specific ventilation with free-breathing hydrogen 1 (¹H) magnetic resonance (MR) imaging without exogenous contrast material and to investigate correlations with hyperpolarized helium 3 (³He) MR imaging and pulmonary function test measurements in healthy volunteers and patients with asthma. Materials and Methods Subjects underwent free-breathing ¹H and static breath-hold hyperpolarized ³He MR imaging as well as spirometry and plethysmography; participants were consecutively recruited between January and June 2017. Free-breathing ¹H MR imaging was performed with an optimized balanced steady-state free-precession sequence; images were retrospectively grouped into tidal inspiration or tidal expiration volumes with exponentially weighted phase interpolation. MR imaging volumes were coregistered by using optical flow deformable registration to generate ¹H MR imaging-derived specific ventilation maps. Hyperpolarized ³He MR imaging- and ¹H MR imaging-derived specific ventilation maps were coregistered to quantify regional specific ventilation within hyperpolarized ³He MR imaging ventilation masks. Differences between groups were determined with the Mann-Whitney test and relationships were determined with Spearman (ρ) correlation coefficients. Statistical analyses were performed with software. Results Thirty subjects (median age: 50 years; interquartile range [IQR]: 30 years), including 23 with asthma and seven healthy volunteers, were evaluated. Both ¹H MR imaging-derived specific ventilation and hyperpolarized ³He MR imaging-derived ventilation percentage were significantly greater in healthy volunteers than in patients with asthma (specific ventilation: 0.14 [IQR: 0.05] vs 0.08 [IQR: 0.06], respectively, P < .0001; ventilation percentage: 99% [IQR: 1%] vs 94% [IQR: 5%], P < .0001). For all subjects, ¹H MR imaging-derived specific ventilation correlated with plethysmography-derived specific ventilation (ρ = 0.54, P = .002) and hyperpolarized ³He MR imaging-derived ventilation percentage (ρ = 0.67, P < .0001) as well as with forced expiratory volume in 1 second (FEV₁) (ρ = 0.65, P = .0001), ratio of FEV₁ to forced vital capacity (ρ = 0.75, P < .0001), ratio of residual volume to total lung capacity (ρ = -0.68, P < .0001), and airway resistance (ρ = -0.51, P = .004). ¹H MR imaging-derived specific ventilation was significantly greater in the gravitational-dependent versus nondependent lung in healthy subjects (P = .02) but not in patients with asthma (P = .1). In patients with asthma, coregistered ¹H MR imaging specific ventilation and hyperpolarized ³He MR imaging maps showed that specific ventilation was diminished in corresponding ³He MR imaging ventilation defects (0.05 ± 0.04) compared with well-ventilated regions (0.09 ± 0.05) (P < .0001). Conclusion ¹H MR imaging-derived specific ventilation correlated with plethysmography-derived specific ventilation and ventilation defects seen by using hyperpolarized ³He MR imaging. ^© RSNA, 2018 Online supplemental material is available for this article.

Holistic segmentation of the lung in cine MRI

Duchenne muscular dystrophy (DMD) is a childhood-onset neuromuscular disease that results in the degeneration of muscle, starting in the extremities, before progressing to more vital areas, such as the lungs. Respiratory failure and pneumonia due to respiratory muscle weakness lead to hospitalization and early mortality. However, tracking the disease in this region can be difficult, as current methods are based on breathing tests and are incapable of distinguishing between muscle involvements. Cine MRI scans give insight into respiratory muscle movements, but the images suffer due to low spatial resolution and poor signal-to-noise ratio. Thus, a robust lung segmentation method is required for accurate analysis of the lung and respiratory muscle movement. We deployed a deep learning approach that utilizes sequence-specific prior information to assist the segmentation of lung in cine MRI. More specifically, we adopt a holistically nested network to conduct image-to-image holistic training and prediction. One frame of the cine MRI is used in the training and applied to the remainder of the sequence ([Formula: see text] frames). We applied this method to cine MRIs of the lung in the axial, sagittal, and coronal planes. Characteristic lung motion patterns during the breathing cycle were then derived from the segmentations and used for diagnosis. Our data set consisted of 31 young boys, age [Formula: see text] years, 15 of whom suffered from DMD. The remaining 16 subjects were age-matched healthy volunteers. For validation, slices from inspiratory and expiratory cycles were manually segmented and compared with results obtained from our method. The Dice similarity coefficient for the deep learning-based method was [Formula: see text] for the sagittal view, [Formula: see text] for the axial view, and [Formula: see text] for the coronal view. The holistic neural network approach was compared with an approach using Demon's registration and showed superior performance. These results suggest that the deep learning-based method reliably and accurately segments the lung across the breathing cycle.

Ultrashort echo time MRI biomarkers of asthma

PURPOSE

To develop and assess ultrashort echo-time (UTE) magnetic resonance imaging (MRI) biomarkers of lung function in asthma patients.

MATERIALS AND METHODS

Thirty participants including 13 healthy volunteers and 17 asthmatics provided written informed consent to UTE and pulmonary function tests in addition to hyperpolarized-noble-gas 3T MRI and computed tomography (CT) for asthmatics only. The difference in MRI signal-intensity (SI) across four lung volumes (full-expiration, functional-residual-capacity [FRC], FRC+1L, and full-inspiration) was determined on a voxel-by-voxel basis to generate dynamic proton-density (DPD) maps. MRI ventilation-defect-percent (VDP), UTE SI, and DPD values as well as CT radiodensity were determined for whole lung and individual lobes.

RESULTS

Mean SI at full-expiration (P < 0.01), FRC (P < 0.05), and DPD (P < 0.01) were greater in healthy volunteers compared to asthmatics. In asthmatics, UTE SI at full-expiration and DPD were correlated with FEV₁ /FVC (SI r = 0.73/P = 0.002; DPD r = 0.75/P = 0.003), RV/TLC (SI r = -0.57/P = 0.02), or RV (DPD r = -0.62/P = 0.02), CT radiodensity (SI r = 0.83/P = 0.006; DPD r = 0.71/P = 0.01), and lobar VDP (SI r_s  = -0.33/P = 0.02; DPD r_s  = -0.47/P = 0.01).

CONCLUSION

In patients with asthma, UTE SI and dynamic proton-density were related to pulmonary function measurements, whole lung and lobar VDP, as well as CT radiodensity. Thus, UTE MRI biomarkers may reflect ventilation heterogeneity and/or gas-trapping in asthmatics using conventional equipment, making this approach potentially amenable for clinical use.

LEVEL OF EVIDENCE

2 J. Magn. Reson. Imaging 2017;45:1204-1215.