Large overlap of "AP" and "CHEST (PA and LAT)" images in MIMIC-CXR

[maximek3]

Prerequisites

• [X] Put an X between the brackets on this line if you have done all of the following:
  • Checked the online documentation: https://mimic.mit.edu/
  • Checked that your issue isn't already addressed: https://github.com/MIT-LCP/mimic-code/issues?utf8=%E2%9C%93&q=

Description

Hi, I have found that there is an overlap of 22,043 chest X-rays in MIMIC-CXR that have as ViewPosition AP but their PerformedProcedureStepDescription is given as CHEST (PA and LAT). Are those X-rays all the ones that are AP, but not taken with a portable scanner?

Thanks for your help!

[alistairewj]

Hmm interesting. I would instinctively trust the ViewPosition - it's required to be present for all the images. No guarantees as I haven't investigated that subset of images in detail.

[maximek3]

Thanks for getting back on this.

It seems that >99% of images have either CHEST (PA and LAT) or CHEST (PORTABLE AP) as value for ProcedureCodeSequence_CodeMeaning.

Would it be fair to assume that whenever an image is both AP and CHEST (PA and LAT), it's an AP scan taken with a fixed scanner? And when it's AP and CHEST (PORTABLE AP), it's an AP scan taken with a portable scanner?

[alistairewj]

That's what I would do. And then I'd check ~10-20 records to make sure it's consistent with what I see. For a while we had a model which classified images as PA or AP, since it's a very easy classification problem, so that's an alternative option.

[maximek3]

Thanks for the pointers, I did this, and here is what I found:

For 20 studies that have both AP and CHEST (PA and LAT):

• 3 mention in their report they are PA scans. But their viewpoint is given as AP. The study ids are 55653823, 58957089, and 59686004. However, when looking at the images, at least for 55653823 and 59686004, it becomes clear that it is actually an AP image (it's marked on the scan), so it seems the comment in the report is wrong, and the viewpoint from the meta file is correct.
• 6 are described as "frontal" in the report.
• The remaining 11 are described as "AP" in the report.

For 10 studies that have AP and CHEST (PORTABLE AP), 8 clearly indicate in the report that the scans were taken on a portable scanner. For the remaining 2 there is no information excluding this.

All in all, it does seem like the assumptions below is correct :)

Would it be fair to assume that whenever an image is both AP and CHEST (PA and LAT), it's an AP scan taken with a fixed scanner? And when it's AP and CHEST (PORTABLE AP), it's an AP scan taken with a portable scanner?