Reader Comments

Post a new comment on this article

Difference in density differentials (fig 4) not supported by fig 2?

Posted by SaraMuller on 17 Jul 2008 at 09:43 GMT

I have 3 comments on the consistency of figure 2, figure 4 and the text:

A: Figure 2 shows the density in all violins with an absolutely calibrated scale. The large difference in density differentials in the top plates between old and new violins as reported in figure 4 should be visible as a larger variation in the older violins (within the ROI regions of figure 3). However, using the ROI's of figure 3 and the original TIFF image no systematic difference between the top plate of old and new violins is visible. The TIFF image is probably lower in resolution than the orignial CT image (I estimate a factor of about 2 compared to the 512 pixel transverse image size, in the other direction the image is oversampled so the difference may be larger), but it seems unlikely that this would remove all the difference between the violins. A possibly explanation would be that the figures of all violins have been individually scaled before being added together. This would invalidate the absolute scale as shown. Could the authors perhaps supply the original data shown in figure 2, and also representative examples of the full shape of the histograms which underlie figure 4?

If indeed lowering the resolution removes the differences between the old and new violins, this means that the differences are only present at a very small scale!

B The results mention a modern violin with extremely low values for both top and back plate. In the top plates one modern top plate with a low density differential is visible, but all back plates have similar values, which is inconsistent with the all wood of this violin being treated.

C The top plate of the middle lower violin shows very density areas in ROI region d and e (and also above region a and b). Why is this not considered a repair, and why are regions d and e not chosen slightly higher (just below the streak artifact), and perhaps slightly more elongated?

Dr Sara H Muller, physicist medical imaging, Amsterdam, The Netherlands

RE: Difference in density differentials (fig 4) not supported by fig 2?

terryborman replied to SaraMuller on 29 Jul 2008 at 09:03 GMT

Hello Dr. Muller,

Thank you for your comments on our paper. It would be best for Berend Stoel to respond to your questions A and B regarding the quantitative analysis but unfortunately he's away from his office for another week or so. As for your C we did state "Due to the increased repair work on one of the classical instruments, it was necessary to choose the ROI's carefully so as to reflect the true wood density, not that of the repair." unfortunately we were obviously not clear enough that the ROI's for all other instruments were the same but for the central violin lower row it was necessary to slightly move these ROI's to remove any repair from the quantitative analysis. I'm sorry for the confusion. Dr. Stoel will hopefully be back and able to address your other questions quite soon.

terry borman

RE: Difference in density differentials (fig 4) not supported by fig 2?

bcstoel replied to SaraMuller on 23 Aug 2008 at 12:46 GMT

Dear Dr Muller,
Thank you for your critical remarks and careful visual analysis of the resulting images.
As for remark A, I am not sure if I understand your remark fully: the density variations within the older violins were actually smaller, not larger. In close-up, this can be seen especially in the top plates of Figure 2 in the two rightmost older violins, although indeed the resolution of these figures may not be sufficient to visually appreciate this fully. Furthermore, one needs to realize that visual perception of intensity differences are very much influenced by the surrounding image structures (for example the well-known Mach-bands would illustrate this). As a result visual interpretation may not correspond completely to the actually measured values. Additionally, we found that the median density measurement of wood was influenced by the thickness of the plate. Therefore, we corrected these measurements, as described in the paper. The images in Figure 2 are the original images in which this influence is still visible. Possibly the most important item to remember when viewing these images is that our analysis was quantitative, not visual, so these images are secondary; our findings were a statistical interpretation of the individual ROI's chosen for each instrument.
The resolution of the individual images is exactly the same, because we used an identical image acquisition protocol, and the CT image were analyzed in a batch job, without human interaction.
On remark B; Our work was focused primarily on the top plates since they are most likely responsible for a higher percentage of the total sound radiation of a violin; that withstanding, the back plates show two "clusters", overlapping to the same degree as the top plates, although on a much smaller scale. The one modern back that is of lowest density was possibly treated. Again, as we stated in our results we don't know anything other than what we found and conjecture as to what may have caused these differences was not within the purview of our work.
Best regards,
Berend Stoel

RE: RE: Difference in density differentials (fig 4) not supported by fig 2?

SaraMuller replied to bcstoel on 07 Sep 2008 at 11:14 GMT

Dear dr Stoel, beste Berend

Trying to keep my comment short I did not fully describe my analysis of your data. I am aware of the dangers of visual interpretation (which I minimized by looking at a wide ranges of zoom factors), so I used Acrobat Photoshop 2 on the TIFF image to create histograms at approximately the same positions as you described in you article (leading to my comment nr 3, which has been resolved satisfactorily by Terry Borman) to see whether I could understand my visual observation.
However, in these histograms I could not see a difference between the old and new violins. And yes you are right, larger is not the correct word, but the problem is that I do not see a difference. Also, I have understood your correction, and I feel this is a proper correction, which moreover cannot be the cause of my problem.
CS2 only yields a numerical estimate for the mean and standard deviation, not a value for the 10-90% value, but looking at the histogram shapes I do not see a difference either.
Therefore my question whether you could supply (examples of) the histograms as you observed them? As you explain that these histogram ROI’s are the basis of your article rather than the figures which you call secondary, I feel that is is it proper to show them.
My remark about combining the images did not concern the resolution, but the grey scale axis: I wondered whether automatic scaling might have occurred on that axis when creating the combined image, leading to my problem even though the original data are correct.
Do I understand from you answer that the TIFF image is the original image? That would surprise me, as counting the pixels and using a typical violin size yields a resolution roughly twice as bad as the resolution stated in the article.

As to remark B:
The remark in the results I refer to is “ As we noticed the one modern top and back plate of extremely extremely low differential”. I do see what you mean by two groups in the back plates, but I do not see an outlier on the differential axis, and certainly not the violin with the lowest density. Therefore, your argument that this violin supports that treated wood has lower densities is not as strong as stated. Of course, it might be that the supplier had only treated the front and not the back plate…, we will never know.

with kind regards, Sara Muller

RE: RE: RE: Difference in density differentials (fig 4) not supported by fig 2?

bcstoel replied to SaraMuller on 25 Nov 2008 at 01:30 GMT

Dear dr Muller, beste Sara

As it is apparently not possible to upload any graphical objects to this discussion site, I have emailed my response directly to you some time ago. There I included two sample images and their histograms from a typical region in a contemporary and classical violin, in which the differences can be appreciated easily. The illustration in the paper has a two-fold lower resolution as the original matlab images. This down-scaling causes degraded histograms in which the differences cannot be seen. For this same reason, actually, we optimized our image acquisition protocol to obtain the most sensitive resolution.

Best regards,

Berend Stoel