Fix profile penumbra part deux #327

randlet · 2020-11-20T19:45:49Z

This is a replacement for PR #324 and includes a fix for LeedsTOR which broke due to improper peak detection.

Warning: I don't 100% understand the difference between threshold & x in find_fwxm_peaks and I need to investigate more whether it breaks anything else. Any thoughts?

This patch resolves issue jrkerns#323 by switching to detecting the outermost points that cross the threshold value when detecting field edges in SingleProfile.

Using default x=50 for find_fwxm_peaks in LeedsTOR caused the wrong peak to be detected. (Don't understand 100% why :?)

randlet · 2020-11-20T20:40:59Z

Not sure why yet but this PR breaks the starshot demo.

jrkerns · 2020-11-20T20:47:30Z

Let me know if these help:

pylinac/pylinac/core/profile.py

Lines 616 to 619 in 265ca6a

    
                   Parameters 
        
                   ---------- 
        
                   x : int, float 
        
                       The Full-Width-X-Maximum desired. E.g. 0.7 will return the FW70%M.

and

pylinac/pylinac/core/profile.py

Lines 558 to 564 in 265ca6a

    
                   threshold : int, float 
        
                       The value the peak must be above to be considered a peak. This removes "peaks" 
        
                       that are in a low-value region. 
        
                       If passed an int, the actual value is the threshold. 
        
                       E.g. when passed 15, any peak less with a value <15 is removed. 
        
                       If passed a float, it will threshold as a percent. Must be between 0 and 1. 
        
                       E.g. when passed 0.4, any peak <40% of the maximum value will be removed.

randlet · 2020-11-20T20:49:16Z

I read those but somehow confused myself. Yes that makes sense. Thanks.

randlet · 2020-11-20T20:50:21Z

With this PR the two outside peaks don't get detected correctly.

jrkerns · 2020-11-20T20:56:00Z

Not sure why yet but this PR breaks the starshot demo.

Looking at the starshot both with your recent change and this change now. I'm also doing some small refactoring.

jrkerns · 2020-11-20T20:58:10Z

FYI the multiple image combine test has been failing for a while.

jrkerns · 2020-11-20T21:20:59Z

Problem is somewhere in here:

pylinac/pylinac/core/profile.py

Lines 641 to 646 in 265ca6a

    
           original_peaks = copy.deepcopy(self.peaks) 
        
           for num, (peak, profile) in enumerate(zip(self.peaks, subprofiles)): 
        
               shift = original_peaks[num - 1].idx if num > 0 else 0 
        
               # shift = sum(len(profile.values) for profile in subprofiles[:num]) 
        
               fwhmc = profile.fwxm_center(x, interpolate=interpolate) 
        
               peak.idx = fwhmc + shift

randlet · 2020-11-20T21:26:28Z

Just came to exact same conclusion! The first subprofile has this shape:

and the new penumbra point detection method is picking up the right hand side. Should the penumbra detection take the gradient of the profile into consideration? So when looking for a profile from RIGHT it should only consider points where gradient is negative?

jrkerns · 2020-11-20T21:27:04Z

So the new code is finding the FWXM based on the indices found, and when you have a profile like this then it will use the furthest-out index. So the FWXM has a right side of 4800.

randlet · 2020-11-20T21:27:18Z

lol

jrkerns · 2020-11-20T21:27:57Z

Just came to exact same conclusion! The first subprofile has this shape:

and the new penumbra point detection method is picking up the right hand side. Should the penumbra detection take the gradient of the profile into consideration? So when looking for a profile from RIGHT it should only consider points where gradient is negative?

I've not thought of using the sign of the gradient. That's pretty smart. I think this is why I did the center-looking-outward approach.

jrkerns · 2020-11-20T21:29:13Z

In the find_peaks function there's a parameter search_region. This helps cut the left and right sides of the profile to avoid these cases. Unfortunately, I've not implemented this for FWXM methods.

jrkerns · 2020-11-20T21:29:59Z

pylinac/pylinac/core/profile.py

Lines 572 to 575 in 265ca6a

    
                   search_region : tuple of ints, floats, or both 
        
                       The region within the profile to search. The tuple specifies the (left, right) edges to search. 
        
                       This allows exclusion of edges from the search. If a value is an int, it is taken as is. If a float, must 
        
                       be between 0 and 1 and is the ratio of the profile length. The left value must be less than the right.

jrkerns · 2020-11-20T21:34:43Z

Hacky, but the smallest solution instead of passing through search_region everywhere is to cut off the left and right ends within subdivide to say, 80% width:

pylinac/pylinac/core/profile.py

Lines 670 to 672 in 265ca6a

    
           left_end = peaks[idx].idx 
        
           peak_idx = peaks[idx+1].idx - left_end 
        
           right_end = peaks[idx+2].idx

randlet · 2020-11-20T21:37:48Z

Check commit I just pushed

…orgot to include

jrkerns · 2020-11-20T22:25:02Z

pylinac/planar_imaging.py

@@ -738,7 +738,7 @@ def _phantom_angle_calc(self) -> float:

        start_angle_deg = self._determine_start_angle_for_circle_profile()
        circle = self._circle_profile_for_phantom_angle(start_angle_deg)
-        peak_idx = circle.find_fwxm_peaks(threshold=0.6, max_number=1)[0]
+        peak_idx = circle.find_fwxm_peaks(threshold=0.6, x=80, max_number=1)[0]


This x=80 is failing a number of tests. Why are you choosing 80?

No good reason...I think that I was somewhat confused and was experimenting with various parameters. (I'm not proud ☹️ )

jrkerns · 2020-11-22T16:51:21Z

I think scipy copied me 😂: https://docs.scipy.org/doc/scipy/reference/generated/scipy.signal.find_peaks.html#scipy.signal.find_peaks

This is new in v1.1, which wasn't around when I created this. I'd much prefer to replace my implementation with theirs. @randlet are you okay with a scipy>=1.1 requirement?

randlet · 2020-11-22T17:02:05Z

Yes that's ok with me.

…

On Sun., Nov. 22, 2020, 11:51 James Kerns, ***@***.***> wrote: I think scipy copied me 😂: https://docs.scipy.org/doc/scipy/reference/generated/scipy.signal.find_peaks.html#scipy.signal.find_peaks This is new in v1.1, which wasn't around when I created this. I'd much prefer to replace my implementation with theirs. @randlet <https://github.com/randlet> are you okay with a scipy>=1.1 requirement? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#327 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAATCMGSKOEYAZGLWCZRBUTSRE6RLANCNFSM4T5F3R4A> .

jrkerns · 2020-11-22T20:11:50Z

Awesome. This is perfect. It will find peaks and also the penumbra points. Refactor....

randlet · 2020-11-22T22:23:16Z

Always nice when you can eliminate a heap of code :)

…ve threshold Previous iteration of _penumbra_point detection failed if there were multiple values above threshold in the profile. This commit ensures the point closest to the peak is used.

randlet · 2020-11-23T17:35:08Z

OK, I went back and actually figured out why the code wasn't working instead of sticking in an arbitrary x=80 🙄 . Thank you for calling me out on that!

The issue was the penumbra point detection still picked the wrong location if there were multiple places in the profile with the correct gradient and the correct threshold. I fixed it so that the location closest to the peak meeting those conditions is used and it all seems good now! This is maybe just academic at this point if you're factoring most of this code out now, but I needed a fix in the interim.

jrkerns · 2020-11-24T02:08:15Z

I am refactoring it 🤷‍♂️. The scipy implementation is very good.

randlet · 2020-11-24T02:36:54Z

See if you can talk them into implementing phantom angle detection next 😆 !

jrkerns · 2020-11-24T03:36:42Z

time to launch a Kaggle contest for it! 😂

SimonBiggs · 2020-11-24T05:08:37Z

See if you can talk them into implementing phantom angle detection next 😆 !

If there's interest in applying a similar logic for phantom angle detection, here is how I detect field image rotation for WLutz work:

https://github.com/pymedphys/pymedphys/blob/3738e899c9944bb592079ce017c04e07abd78226/lib/pymedphys/_wlutz/findfield.py#L197-L220

SimonBiggs · 2020-11-24T10:04:03Z

See if you can talk them into implementing phantom angle detection next laughing !

Probably a better and more clear example of using scipy's basinhopping to determine rotation:

https://github.com/pymedphys/pymedphys/blob/3738e899c9944bb592079ce017c04e07abd78226/lib/pymedphys/_electronfactors/core.py#L384-L422

randlet · 2020-11-24T14:24:18Z

Thanks! I'm implementing a PTW EPID QC phantom now so maybe I'll give that a shot.

) RAM-3200 Add refinement for 604 analyses Approved-by: Randy Taylor

randlet added 2 commits November 20, 2020 14:35

Use outermost threshold crossing point when detecting field edges

648ebc4

This patch resolves issue jrkerns#323 by switching to detecting the outermost points that cross the threshold value when detecting field edges in SingleProfile.

Fix Leeds TOR phantom angle detection with new penumbra_point method

de37a00

Using default x=50 for find_fwxm_peaks in LeedsTOR caused the wrong peak to be detected. (Don't understand 100% why :?)

use gradient to qualify penumbra diretion

d884c12

add back the check for case where no penumbra point is found that I f…

ea7d71e

…orgot to include

jrkerns reviewed Nov 20, 2020

View reviewed changes

randlet added 2 commits November 23, 2020 12:23

Fix _penumbra_point detection if profile contains multiple values abo…

17afc57

…ve threshold Previous iteration of _penumbra_point detection failed if there were multiple values above threshold in the profile. This commit ensures the point closest to the peak is used.

remove hacky fw80m now that _penumbra_point picks correct penumbra index

ef5fb1a

jrkerns closed this Feb 9, 2021

jrkerns added a commit that referenced this pull request Jan 11, 2024

Merged in bugfix/RAM-3200_catphan_604_slice_thickness (pull request #327

d8ea67d

) RAM-3200 Add refinement for 604 analyses Approved-by: Randy Taylor

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fix profile penumbra part deux #327

Fix profile penumbra part deux #327

randlet commented Nov 20, 2020

randlet commented Nov 20, 2020

jrkerns commented Nov 20, 2020

randlet commented Nov 20, 2020

randlet commented Nov 20, 2020

jrkerns commented Nov 20, 2020

jrkerns commented Nov 20, 2020

jrkerns commented Nov 20, 2020

randlet commented Nov 20, 2020

jrkerns commented Nov 20, 2020

randlet commented Nov 20, 2020

jrkerns commented Nov 20, 2020

jrkerns commented Nov 20, 2020

jrkerns commented Nov 20, 2020

jrkerns commented Nov 20, 2020

randlet commented Nov 20, 2020

jrkerns Nov 20, 2020

randlet Nov 20, 2020

jrkerns commented Nov 22, 2020

randlet commented Nov 22, 2020 via email

jrkerns commented Nov 22, 2020

randlet commented Nov 22, 2020

randlet commented Nov 23, 2020

jrkerns commented Nov 24, 2020

randlet commented Nov 24, 2020

jrkerns commented Nov 24, 2020

SimonBiggs commented Nov 24, 2020

SimonBiggs commented Nov 24, 2020

randlet commented Nov 24, 2020

Fix profile penumbra part deux #327

Fix profile penumbra part deux #327

Conversation

randlet commented Nov 20, 2020

randlet commented Nov 20, 2020

jrkerns commented Nov 20, 2020

randlet commented Nov 20, 2020

randlet commented Nov 20, 2020

jrkerns commented Nov 20, 2020

jrkerns commented Nov 20, 2020

jrkerns commented Nov 20, 2020

randlet commented Nov 20, 2020

jrkerns commented Nov 20, 2020

randlet commented Nov 20, 2020

jrkerns commented Nov 20, 2020

jrkerns commented Nov 20, 2020

jrkerns commented Nov 20, 2020

jrkerns commented Nov 20, 2020

randlet commented Nov 20, 2020

jrkerns Nov 20, 2020

Choose a reason for hiding this comment

randlet Nov 20, 2020

Choose a reason for hiding this comment

jrkerns commented Nov 22, 2020

randlet commented Nov 22, 2020 via email

jrkerns commented Nov 22, 2020

randlet commented Nov 22, 2020

randlet commented Nov 23, 2020

jrkerns commented Nov 24, 2020

randlet commented Nov 24, 2020

jrkerns commented Nov 24, 2020

SimonBiggs commented Nov 24, 2020

SimonBiggs commented Nov 24, 2020

randlet commented Nov 24, 2020