Mapping Alveolar Oxygen Partial Pressure in COPD Using Hyperpolarized Helium-3: The Multi-Ethnic Study of Atherosclerosis (MESA) COPD Study

Round 1

Reviewer 1 Report

The authors describe an in-depth study investigating a subgroup of the MESA cohort with 3He MRI as well as standard pulmonary function measures. From the 3He MRI, the primary measure was alveolar oxygen partial pressure described by mean, SD, and % non-physiological. Comparisons were completed with respect to other 3He measurements, CT measurements, and PFT measurements. The main goal was to investigate the usefulness and application of 3HE paO2 to COPD patients.

Overall, the experiments were well thought out and scientifically sound and the paper was well written. This is a larger study for HP MRI on an important disease population. In my opinion, only minor corrections and clarifications are necessary for publication.

1.      Page 2, Line 70: The “HP no-ble gas MRI directly” sentence seems to repeat line 67

2.      Page 2, Line 90-97: This paragraph seems to jump around on topics and left me more confused and unsure how, and if, these are linked. For example, what emphysema subtypes classification? Using paO2 or CT or ADC or PFTs? If novel, it should have a reference. Also, why does the number of participants and disease severity matter for paO2 or emphysema subtypes in this study?

3.      Table 1: I personally find percent predicted, rather than volume more informative about disease severity. (FEV1, FVC)

4.      Page 4, Line 139: The text here doesn’t line up with Table 2 for breath-hold level for paO2. Please clarify/fix.

5.      Page 5, Line 149: When/how was the proton MRI collected? Between acquisitions? Before or after 3He? Interleaved with a 3He acq?

6.      Page 5, Line 15: I think this should be “after n acquisitions” like in line 156 since that counts towards your total number of RF excitations.

7.      Page 6, Line 182-187: The thresholds described here do not make sense to me. Is Sheart supposed to represent noise sampled from the heart region of the 3He image? And then nonventilated <= Sheart mean + 2.6 x Sheart SD? Is Strachea supposed to represent max signal so that fully ventilated is >= noise + 25% of (max signal – mean signal). The rest being hypo?

8.      Page 8, Line 252 and Sup Fig A: How was no usable images defined? What were the reasons for the 2 paO2 and 6 ADC incompletions?

9.      Page 8, Line 257: I think with should be without in “subject with COPD presents”

10.  Fig 3: It would be nice to see trend lines to the data to better visualize the correlations

11.  Page 14, Line 397: Could you elaborate more on what imaging artifacts discussed here that is worse at higher fields and makes the data unusable but can be corrected for and the data recovered?

12.  Page 14, Line 417-419: What are “these regions”? Regions of negative paO2? Delayed ventilation regions? The first part makes me think delayed ventilation, but the second part make me think paO2. And thus, the claim HP MRI provides insight into earlier stages doesn’t make sense

13.  Page 14, Line 431: Should “non” be “none”?

Author Response

Thank you very much for your comments and suggestions. We have implemented the changes you suggested and we have also re-formatted the tables so they would fit within the margins.

Here are our answers to your questions:

1. Page 2, Line 70: The “HP no-ble gas MRI directly” sentence seems to repeat line 67

We deleted the sentence in line 70, added the word regional to line 67.

2. Page 2, Line 90-97: This paragraph seems to jump around on topics and left me more confused and unsure how, and if, these are linked. For example, what emphysema subtypes classification? Using paO2 or CT or ADC or PFTs? If novel, it should have a reference. Also, why does the number of participants and disease severity matter for paO2 or emphysema subtypes in this study?

Reference 4 was also added after the section on emphysema subtypes as the novel method is described in reference 4, which describes the emphysema subtypes classification. We are studying the sensitivity of p_AO₂ to emphysema subtypes determined in six zones using CT. We re-worded information in introduction and added reference but see paragraph lines 217-224 for more details.

3. Table 1: I personally find percent predicted, rather than volume more informative about disease severity. (FEV1, FVC)

FEV1 and FVC are given in units of volume and characterize the subject population, they are not compared to expectations.

4. Page 4, Line 139: The text here doesn’t line up with Table 2 for breath-hold level for paO2. Please clarify/fix.

The total volume in all cases of helium plus nitrogen is 1 L.

5. Page 5, Line 149: When/how was the proton MRI collected? Between acquisitions? Before or after 3He? Interleaved with a 3He acq?

The 5 s 3He MRI was acquired first. It took ~3 s to switch the scanner from 3He to 1H acquisition. A 6 s 1H MRI scan was then acquired afterwards within the same breath hold.

6. Page 5, Line 15: I think this should be “after n acquisitions” like in line 156 since that counts towards your total number of RF excitations.

Could you clarify which line you are referring to?

7. Page 6, Line 182-187: The thresholds described here do not make sense to me. Is Sheart supposed to represent noise sampled from the heart region of the 3He image? And then nonventilated <= Sheart mean + 2.6 x Sheart SD? Is Strachea supposed to represent max signal so that fully ventilated is >= noise + 25% of (max signal – mean signal). The rest being hypo?

That is correct, Sheart represents noise and Strachea represents maximum signal. The threshold was set up by image analysis experts in our group. Please see detailed description of the analysis methods below in our recently published paper. We have now updated our reference [39] to include this recently published paper.

We developed the MRI ventilation method leveraging the experience of our Image Analysis Core Lab as well as the combined experience of image analysts, radiologists, physiologists, and physicians. The image analysts have 15-20 years of experience in MRI analysis at Image Analysis Core Lab and the radiologist has over 35 years of experience in MRI reading and evaluation.

Full lung mask annotation: Lung masks were segmented on coronal ¹H MRI images of each participant to define the lung boundaries (Figure 1). A region of interest (ROI) was drawn on each coronal slice within each lung excluding regions with partial volume effects (i.e., loss of contrast between two adjacent tissues in a slice caused by insufficient resolution so that more than one tissue type occupies the same voxel or pixel). The mean and SD of the signal intensity (SI) of the lung ROI were calculated for each coronal slice of lungs. The lungs were then segmented on ¹H MRI images using a region growing method with threshold values adaptively tied to mean SI_{lung ROI} ± 2.576 SD_{lung ROI} (Figure 1). Manual corrections were applied if necessary. The corrections were mostly small anatomical corrections such as removing low intensity regions including cortical bones, central airways and background.

Ventilation annotation: The full lung masks from ¹H MRI images were applied to each corresponding ³He slice to define the lung boundaries on ³He MRI (Figure 1). SI of heart was used as a reference for non-ventilated regions and SI of central airways was used as a reference for ventilated regions. ROIs of the heart and central airways were manually defined on the ³He MRI images excluding regions with partial volume effect. Non-ventilated regions were segmented via region growing with a threshold defined as the mean SI_{heart ROI} ± 2.576 SD_{heart ROI}. After non-ventilated regions were annotated, ventilated regions were further divided into normal-ventilated regions and hypo-ventilated regions. Normal-ventilated regions were segmented by region growing with a threshold as mean SI_{heart ROI} + 0.25 (mean SI_{central airway ROI} – mean SI_{heart ROI}). The remaining lung regions (i.e., the first quartile of the range of mean SI_{heart ROI} and mean SI_{central airway ROI}) were defined as hypo-ventilated regions (Figure 1).

The analysts were blinded to the clinical information of the participants. One analyst performed the image analysis. A second analyst was included to evaluate inter-reader agreement. Two readings were included for intra-reader agreement. A total of 10 randomly selected MRI scans were used the intra-reader and inter-reader agreement assessment and were read twice. The intra-reader and inter-reader % CV was 1.7% and 2.6% for non-ventilated, 3.8% and 3.4% for hypo-ventilated, and 3.4% and 4.0% for normal-ventilated lung regions, respectively. SliceOmatic (TomoVision, Magog, Canada) was used to perform the manual ROI selection, region growing and manual corrections.

See attached image (which is not included in the manuscript but is attached here for additional clarification).

8. Page 8, Line 252 and Sup Fig A: How was no usable images defined? What were the reasons for the 2 paO2 and 6 ADC incompletions?

“no usable images” was defined as only noised on the image. The lost ADC data was due to failed “breath-holds”

9. Page 8, Line 257: I think with should be without in “subject with COPD presents”

Yes, we changed it to “without”.

10. Fig 3: It would be nice to see trend lines to the data to better visualize the correlations.

We have added the trend lines.

11. Page 14, Line 397: Could you elaborate more on what imaging artifacts discussed here that is worse at higher fields and makes the data unusable but can be corrected for and the data recovered?

MR Imaging artifacts occasional appeared in the Phillips 3T scanner, but were straightforward to identify and remove during the analysis.

 

12. Page 14, Line 417-419: What are “these regions”? Regions of negative paO2? Delayed ventilation regions? The first part makes me think delayed ventilation, but the second part make me think paO2. And thus, the claim HP MRI provides insight into earlier stages doesn’t make sense

“These regions” refer to negative p_AO₂ regions. Negative p_AO₂ studies in mild COPD cases had not been observed before, they had only been seen in severe COPD studies.  Negative p_AO₂ does provide new information about early stages of the disease since we were able to observe it in milder cases.

13. Page 14, Line 431: Should “non” be “none”?

Yes, we changed it to “none”.

Author Response File: Author Response.docx

Reviewer 2 Report

Dear authors

I enjoyed reading your work and believe it is well suited for Tomography. Your findings in a large cohort are important to determine the validity and significance of hyperpolarized 3He lung imaging. Likewise are your investigation into sub-types and your technical developments with data analysis and benchmark for 129Xe important for further studies.  

I have included my minor comments in the attached marked-up pdf.

Best regards

Comments for author File: Comments.pdf

Author Response

Thank you very much for your comments and suggestions. We have implemented the changes you suggested and we have also re-formatted the tables so they would fit within the margins

Here are our answers to your questions/suggestions:

 

• We changed weight to kg and corrected the height: 175.8 +8 cm rather than 75.8 + 4.8, it was a typo.
• We changed liters to L, seconds to s.
• We removed bold font from the ADC formula.
• Line 279: changed to “table 3”.

 

• Why the 2.6 threshold was chosen and the weight in the formula: from reference 39?

The 2.6 sigma threshold corresponds to a 99% confidence interval assuming a Gaussian distribution of the data.

Please see detailed description of the analysis methods below in our recently published paper. We have now updated our reference [39] to include this recently published paper.

We developed the MRI ventilation method leveraging the experience of our Image Analysis Core Lab as well as the combined experience of image analysts, radiologists, physiologists, and physicians. The image analysts have 15-20 years of experience in MRI analysis at Image Analysis Core Lab and the radiologist has over 35 years of experience in MRI reading and evaluation.

Full lung mask annotation: Lung masks were segmented on coronal ¹H MRI images of each participant to define the lung boundaries (Figure 1). A region of interest (ROI) was drawn on each coronal slice within each lung excluding regions with partial volume effects (i.e., loss of contrast between two adjacent tissues in a slice caused by insufficient resolution so that more than one tissue type occupies the same voxel or pixel). The mean and SD of the signal intensity (SI) of the lung ROI were calculated for each coronal slice of lungs. The lungs were then segmented on ¹H MRI images using a region growing method with threshold values adaptively tied to mean SI_{lung ROI} ± 2.576 SD_{lung ROI} (Figure 1). Manual corrections were applied if necessary. The corrections were mostly small anatomical corrections such as removing low intensity regions including cortical bones, central airways and background.

Ventilation annotation: The full lung masks from ¹H MRI images were applied to each corresponding ³He slice to define the lung boundaries on ³He MRI (Figure 1). SI of heart was used as a reference for non-ventilated regions and SI of central airways was used as a reference for ventilated regions. ROIs of the heart and central airways were manually defined on the ³He MRI images excluding regions with partial volume effect. Non-ventilated regions were segmented via region growing with a threshold defined as the mean SI_{heart ROI} ± 2.576 SD_{heart ROI}. After non-ventilated regions were annotated, ventilated regions were further divided into normal-ventilated regions and hypo-ventilated regions. Normal-ventilated regions were segmented by region growing with a threshold as mean SI_{heart ROI} + 0.25 (mean SI_{central airway ROI} – mean SI_{heart ROI}). The remaining lung regions (i.e., the first quartile of the range of mean SI_{heart ROI} and mean SI_{central airway ROI}) were defined as hypo-ventilated regions (Figure 1).

The analysts were blinded to the clinical information of the participants. One analyst performed the image analysis. A second analyst was included to evaluate inter-reader agreement. Two readings were included for intra-reader agreement. A total of 10 randomly selected MRI scans were used the intra-reader and inter-reader agreement assessment and were read twice. The intra-reader and inter-reader % CV was 1.7% and 2.6% for non-ventilated, 3.8% and 3.4% for hypo-ventilated, and 3.4% and 4.0% for normal-ventilated lung regions, respectively. SliceOmatic (TomoVision, Magog, Canada) was used to perform the manual ROI selection, region growing and manual corrections.

See attached image (which is not included in the manuscript but is attached here for additional clarification).

• “A significant challenge for studying p_AO₂ in moderate/severe COPD participants is the ³He gas mixing and voxel to voxel transport between acquisitions, resulting in negative p_AO₂ ” Does this give bias even for non-negative values?

Yes, gas mixing gives bias even for non-negative values. However, the effects on the overall mean pAO2 are negligible especially in healthy lungs. The distribution of pAO2 values widens but there is no significant effect on the mean.

Author Response File: Author Response.docx