Fig 1.
Oscillation grading scale. Oscillations in the raw RBC signal were graded from 1 (no confidence in visibility of oscillation) to 5 (high confidence in visibility of oscillation).
Fig 2.
Reliability of standard Xe-MRI quantitative metrics.
RBC/Membrane (A) shows the strongest reliability (ICC = 0.98), though all measures, including Membrane/Gas (C), RBC/Gas (E), and RBC Defect Percent (G) have ICC values greater than or equal to 0.82, implying strong reliability of these measures. Bland Altman plots (B, D, F, H) are shown alongside correlation plots to emphasize the good agreement between measures.
Fig 3.
Reliability of Global RBC Oscillation Amplitude for different binning strategies.
When a bandpass filter is not used (A, B, E, F), it is more challenging to accurately identify oscillation peaks and valleys, which leads to worse reliability, even when manual intervention is implemented (ICC = 0.7 for automatic detection, ICC = 0.83 for manual intervention) (E). When a bandpass filter is used, oscillation peaks and valleys are able to be detected, and reliability is strong for both automated (C, D; ICC = 0.87) and manual intervention cases (G, H; ICC = 0.88). Bland Altman plots (B, D, F, H) are shown alongside correlation plots to highlight improved agreement between scans for bandpass filtering methods. Abbreviations: BP – Bandpass Filter.
Fig 4.
Representative RBC oscillation traces for each of the grades 1–5, 1 being the worst quality and 5 being the highest.
Column 1 shows the raw RBC signal and the smoothed oscillation trace. Column 2 shows the Fourier transform of the RBC oscillation, with heart rate (HR) and full width at half maximum (FWHM) provided. Column 3 shows bandpass-filtered RBC oscillations, demonstrating a clear improvement in the ability to resolve oscillation peaks and valleys.
Fig 5.
Comparison of quantitative quality metrics to reader scoring.
(A) Full width at half maximum (FWHM) of the dominant oscillation peak was largest for low scores. (B) Heart rate (HR) difference was similarly largest for low scores. (C) Signal-to-noise ratio was largest for high scores. (D) A composite score composed of these three measures effectively discriminated between the poorest oscillation quality (Score of 1), moderate oscillation quality (Score of 2 or 3), and good oscillation quality (Score of 4 or 5). Symbols signify significant differences (p < 0.05): ¶: versus 1; ‡: versus 2; §: versus 3; #: versus 4.
Fig 6.
Central coronal slices for representative oscillation amplitude maps for the 3 algorithms and 3 participant groups investigated in this study.
For SSc and PAH patients, the two scans that occurred back-to-back on the same day are shown. In the healthy participants, both 2-key methods provide similar results, but regional differences are observed in SSc and PAH patients, likely due to regional variation in RBC signal. Multi-key images show significantly different features and much larger values for oscillation amplitude, owing to the phase information also obtained by that method.
Fig 7.
Central coronal slice from representative participants demonstrating that 2-key voxel-scaled amplitude maps can be largely reproduced using the multi-key data.
Scaling the amplitude by the cosine of the oscillation phase brings the range of values to the same as the 2-key method and largely reproduces regional differences in oscillation amplitude.
Fig 8.
Boxplots showing differences in oscillation amplitude in the anterior vs. posterior (A) and in the apex vs base (B) of the lungs for healthy volunteers.
There is a clear difference in oscillation amplitude between the anterior and posterior of the lungs, though the direction of this difference is opposite for the mean-scaled method as compared to the voxel-scaled and multi-key methods. There is no significant difference in oscillation amplitude between the apex and base of the lungs.
Fig 9.
Boxplots showing the difference in mean oscillation amplitude between healthy, PAH, and SSc participants for the different oscillation mapping techniques investigated: (A) Global, (B) 2-key, mean-scaled; (C) 2-key, voxel-scaled, (D) Multi-key, (E) Multi-key scaling mapped amplitude by the cosine of the mapped phase.
P-values from post-hoc testing are shown above plots, with significant (p < 0.05) values shown in red. The global measure of oscillation amplitude (A) provides the best discrimination between participant groups. Each of the oscillation mapping techniques leads to greater overlap, but, in all cases, the PAH patients exhibit the lowest amplitude and SSc the highest. Scaling multi-key amplitude by the cosine of the phase strengthens the separation between participant groups (E) as compared to just the multi-key amplitude.
Table 1.
Same-day ICC and p values for whole-lung mean oscillation amplitude as well as the percentage of the lungs binned to low, mid, and high values. Each comparison is done for all participants as well as for SSc and PAH participants individually. For PAH patients, the 6-week reliability data is also provided.
Fig 10.
Comparison of regional image comparison metrics for two back-to-back images for the different oscillation mapping techniques investigated.
(A) Structural Similarity index is generally high, with the best values observed for the 2-key, mean-scaled maps. (B) Mean square error is relatively low for the 2-key methods but very large for the multi-key method. This appears to be driven by a small number of significant outliers. (C) The average distance metric shows that there is reasonable overlap between regions binned to low, mid, and high values.