Chronic obstructive pulmonary disease (COPD) is diagnosed and monitored using pulmonary function tests that measure airflow limitation stemming from a combination of airway remodeling and parenchymal destruction. While rapid and inexpensive, spirometry provides only a global measurement of lung function and cannot differentiate underlying pathophysiological determinants of lung function, morbidity and mortality in COPD. As shown in Table 1, pulmonary imaging methods like CT and MRI provide lung disease biomarkers that quantify the various phenotypes and pathophysiological features of COPD. We think that a complementary approach may be required to deeply phenotype COPD patients for therapy decisions and to manage disease progression and exacerbations. Recently, parametric-response-mapping (PRM) has been proposed as a way to evaluate thoracic CT to differentiate the regional contributions of small-airways-disease and emphysema.¹ CT is nearly universally available and can be rapidly acquired making PRM very attractive clinically, but the validation of PRM measurements is yet to be undertaken. In addition, automated COPD patient phenotyping based on the structure-function information provided by a combination of CT and MRI has not been undertaken. Thus, our objective was to develop multimodal-parametric-response-mapping (mPRM) using MRI and CT biomarkers to phenotype COPD patients. We hypothesized that mPRM could be used to differentiate regional underlying pathologies of COPD with greater sensitivity and precision compared to either MRI or CT alone. As a first step towards testing this hypothesis we develop such a multi-parametric approach in relatively large group of 68 ex- and never smokers.

Image Acquisition: Hyperpolarized ³He MRI static ventilation images (total-acquisition-time=10s; TR/TE/flip-angle=3.8ms/1.0ms/7°; FOV=40×40cm; matrix=128×80; BW=62.50kHz; NEX=1; number-of-slices=14; slice-thickness=15mm) and diffusion-weighted images (total-acquisition-time=14s; TR/TE/flip-angle=6.8ms/4.5ms/8°; FOV=40×40cm; matrix=128×80; BW=62.50kHz; NEX=1; number-of-slices=7; slice-thickness=30mm) were acquired system as previously described² on a 3T Discovery MR750 (General Electric Health Care, Milwaukee, Wisconsin, USA). ¹H MRI (total-data-acquisition-time=12s; TR/TE/flip-angle=4.3ms/1.0ms/30°; FOV=40×40cm; matrix=128×128; BW=62.50kHz; NEX=1; number-of-slices=14; slice-thickness=15mm) were acquired as previously described.² CT images (inspiratory/expiratory) were acquired on a 64-slice Lightspeed-VCT scanner (GEHC) using a spiral acquisition approach (detector-configuration=64×0.625mm; tube-voltage=120kVp; tube-current=100mA; tube-rotation-time=500ms; pitch=1.0, slice-thickness=1.25mm; number-of-slices=200-250[patient-size-dependent]; matrix=512×512), as previously described³ and reconstructed using a standard convolution kernel to 1.25mm.

Image Analysis: ³He MRI ADC and VDP were calculated as previously described.⁴ Non-rigid image registration was performed using NiftyReg⁵ to register inspiratory/expiratory CT to the reference ¹H image. Affine registration was also performed using NiftyReg to register ³He MRI to ¹H MRI so that all images were in the same space to perform voxel-wise comparisons. Voxel-pair classifications were generated from co-registering ADC and CT into four distinct groups: Group 1 - ADC voxels <0.3cm²/s and >0.0cm²/s and inspiratory/expiratory CT voxels >=-950HU/-856HU; Group 2 - ADC voxels <0.3cm²/s and >0.0cm²/s and inspiratory/expiratory CT voxels <-950HU/-856HU; Group 3 - ADC voxels >=0.3cm²/s or =0.0cm²/s and inspiratory/expiratory CT voxels >=-950HU/-856HU; Group 4 - ADC voxels >=0.3cm²/s or =0.0cm²/s and inspiratory/expiratory CT voxels <-950HU/-856HU. CT thresholds were determined previously to quantify gas-trapping and emphysema in COPD subjects.¹ ADC threshold of 0.3cm²/s was based on previously determined measurements of ADC in healthy volunteers and mild-to-moderate COPD subjects.² Principal component analysis was performed on the 3-dimensional distribution of voxel-triples produced when performing voxel-wise comparison between CT (either inspiratory/expiratory) and ³He static ventilation and ADC maps.

Figure 1 shows mPRM and voxel distributions generated from co-registered inspiratory or expiratory CT with MRI ADC maps. Distributions of voxel-pairs were more variable as disease severity increased. For the ex-smoker with more advanced GOLD III COPD there was a greater number of mPRM voxels reflective of emphysema. Figure 1 also shows the principal-component-analysis of the voxel distribution generated from co-registered inspiration or expiratory CT with ³He MRI SV cluster maps and ³He MRI ADC maps where the ellipsoids represent the 95% confidence interval. The variance in the distribution is represented by the size of the ellipsoid. As illustrated, variance increased when comparing the ex-smoker with subjects with COPD.In this proof-of-concept demonstration, we generated mPRM from CT and MRI pulmonary measurements and performed principal component analysis of the voxel distribution generated from co-registered inspiration or expiratory CT with ³He MRI SV cluster maps and ³He MRI ADC maps for ex-smokers with and without COPD. Further work is necessary to determine the appropriate combination of imaging biomarkers generated from MRI and CT to provide useful information in deeply phenotyping COPD.