Extending semiautomatic ventilation defect analysis for hyperpolarized (129)Xe ventilation MRI

Abstract

Rationale and objectives: Clinical deployment of hyperpolarized (129)Xe magnetic resonance imaging requires accurate quantification and visualization of the ventilation defect percentage (VDP). Here, we improve the robustness of our previous semiautomated analysis method to reduce operator dependence, correct for B1 inhomogeneity and vascular structures, and extend the analysis to display multiple intensity clusters.

Materials and methods: Two segmentation methods were compared-a seeded region-growing method, previously validated by expert reader scoring, and a new linear-binning method that corrects the effects of bias field and vascular structures. The new method removes nearly all operator interventions by rescaling the (129)Xe magnetic resonance images to the 99th percentile of the cumulative distribution and applying fixed thresholds to classify (129)Xe voxels into four clusters: defect, low, medium, and high intensity. The methods were applied to 24 subjects including patients with chronic obstructive pulmonary disease (n = 8), age-matched controls (n = 8), and healthy normal subjects (n = 8).

Results: Linear-binning enabled a faster and more reproducible workflow and permitted analysis of an additional 0.25 ± 0.18 L of lung volume by accounting for vasculature. Like region-growing, linear-binning VDP correlated strongly with reader scoring (R(2) = 0.93, P < .0001), but with less systematic bias. Moreover, linear-binning maps clearly depict regions of low and high intensity that may prove useful for phenotyping subjects with chronic obstructive pulmonary disease.

Conclusions: Corrected linear-binning provides a robust means to quantify (129)Xe ventilation images yielding VDP values that are indistinguishable from expert reader scores, while exploiting the entire dynamic range to depict multiple image clusters.

Keywords: Hyperpolarized (129)Xe MRI; image registration; lung segmentation; ventilation defect analysis.

Copyright © 2014 AUR. Published by Elsevier Inc. All rights reserved.

Figures

Figure 1

Seeded region-growing segmentation workflow: the 129Xe ventilation image (a) was segmented to obtain a ventilation mask (b). The 1H image was registered to the 129Xe image and segmented to obtain a thoracic cavity mask (c). These masks were multiplied to remove major airways from the ventilation mask, thereby obtaining the ventilated volume within the thoracic cavity, enabling calculation of the binary ventilation defect map (d) as well as ventilation defect percentage.

Figure 2

Corrected linear-binning map workflow. The bias-field corrected ventilation images (a) were used to register the 1H thoracic cavity images (b). These were then segmented and detected vasculature was removed to define the thoracic cavity mask that constrains the analysis (c). Pixels from the bias-field corrected ventilation image lying within the thoracic cavity volume were rescaled by their top percentile to range from 0-1 and classified into 4 intensity clusters to create the linear-binning map (d).

Figure 3

Tuning of the thoracic cavity mask. The thoracic cavity image (a) was segmented and morphologically filled to generate the initial thoracic cavity mask (b). The thoracic cavity image then underwent vesselness filtering to detect the pulmonary vasculature (c). This was then removed from the original thoracic cavity image to generate the final thoracic cavity mask (d).

Figure 4

129Xe MRI histogram re-scaling: the native 129Xe MR image (a) has a histogram with a high-intensity “tail” (d) that must be removed for effective rescaling. If this is done by simply dividing all intensities by their top 5%, the resulting binning maps (b) over-estimate the low-intensity bins because the tail has only been partially removed (e). By instead scaling all 129Xe intensity pixels by the 99th percentile of the 129Xe intensity cumulative distribution, the associated binning maps are more reproducible and more consistent with reader perception (c) because the histogram tail has been effectively removed (f).

Figure 5

Impact of the ‘vesselness’ filter. The thoracic cavity image (a) and 129Xe ventilation image (b) in a healthy volunteer. The thoracic cavity mask after initial segmentation (c) overestimates the large vessels as shown in the union with 129Xe MRI (d). The thoracic cavity mask after morphological closing (e) slightly reduces the over-estimation of the larger vessels, but completely eliminates the exclusion of smaller vasculature from the mask (f). The mask after application of ‘vesselness’ filter (g) now shows a thoracic cavity mask that best excludes the vasculature, while preserving a maximum 129Xe image volume for quantitative analysis (h).

Figure 6

Bias-field correction: Column 1 shows ventilation images from two subjects (coronal and axial views) prior to the application of the retrospective bias correction algorithm. Column 2 depicts the estimated bias-fields in both orientations (shown as a maximum intensity projection). Column 3 shows the 129Xe images after bias-field correction. Bias fields appear most intense near the coil elements.

Figure 7

Ventilation images (prior to bias-field correction) with associated reader-based scores, along with ventilation maps generated by the seeded region-growing and corrected linear-binning methods. Shown here are representative cases of a healthy volunteer, an age-matched control, and a COPD subject.

Figure 8

Ventilation defect measurements for healthy volunteers, age-matched controls, and COPD subjects obtained by expert reader scoring, seeded region-growing, and linear-binning.

Figure 9

Correlation of the expert reader-based VDS% with VDP calculated using the seeded region-growing and linear-binning methods. Note that although both methods correlate well with reader scores, Bland-Altman plots show a systematic bias in seeded region-growing that is significantly reduced for linear binning.

Figure 10

Examples of slight discordance between reader scoring and binning. Subject 1 depicts a healthy volunteer appearing to have no clear defects, but when viewed in the context of the registered thoracic cavity, shows clear ventilation defects and low intensity primarily in the apex of the right lung. The second row shows images of an age-matched control with a somewhat tortuous thoracic cavity, which may have caused readers to assign a higher VDS% of 16.67%. However, binning analysis, which incorporates the tortuous thoracic cavity shows mostly medium intensity.