Error in generate_left_ventricle_segments and get_com functions

[hannahalme]

Hello,

Thank you for this useful library! I have used platipy for calculating the AHA-17 segments in cardiac CT data. I noticed that for some patients, the function generate_left_ventricle_segments fails, giving the following error:

File ~/Desktop/segmentointi/test_env/lib/python3.13/site-packages/platipy/imaging/utils/ventricle.py:397, in generate_left_ventricle_segments(contours, label_left_ventricle, label_left_atrium, label_right_ventricle, label_heart, myocardium_thickness_mm, hole_fill_mm, optimiser_tol_degrees, optimiser_max_iter, min_area_mm2, verbose)
    395 for n in range(inf_limit_lv, apical_extent):
    396     lv_com_apical_list.append(get_com(working_contours[label_left_ventricle][:, :, n]))
--> 397     rv_com_apical_list.append(get_com(working_contours[label_right_ventricle][:, :, n]))
    399 lv_com_apical = np.mean(lv_com_apical_list, axis=0)
    400 rv_com_apical = np.mean(rv_com_apical_list, axis=0)

File ~/Desktop/segmentointi/test_env/lib/python3.13/site-packages/platipy/imaging/label/utils.py:86, in get_com(label, as_int, real_coords)
     84     if as_int:
     85         print(com)
---> 86         com = [int(i) for i in com]
     88 return com

ValueError: cannot convert float NaN to integer

To me, it seems that the get_com function tries to calculate the center of mass slice by slice for the left and right ventricle, but fails since some of the slices in the right ventricle contain only zeros (that's why the center of mass is NaN).

Do you think this error could be resolved by allowing the center of mass to be zero for some slices, or would it break some essential functionality? Does the function assume that left and right ventricle have equal length (from inf_limit_lv to apical_extent)? I suppose this is rarely the case in reality.

Thank you for your help!

Best,
Hanna

[hannahalme]

Hi all,
Have you had a chance to look into this issue yet?
@kirbyju @SimonBiggs @pchlap @rnfinnegan
Thank you very much for your time and effort—I really appreciate it.
Best,
Hanna

[pchlap]

Hi @hannahalme,
@rnfinnegan may be best positioned to answer this for you. If he isn't available then I could try, but I would really need the same data you are using to run this. If you can share the structure data only (ensuring no patient identifiable meta data is present), then I could try to run this on my end and see whats going on.

[hannahalme]

Hi,
Thank you for your reply. Would you prefer that I share the structure data files by email, or should I upload them here in this thread?

[pchlap]

Uploading them to something like OneDrive or Dropbox would be best. Then you can email me the link directly if you don't want them to be public: phillip.chlap@unsw.edu.au

[hannahalme]

@pchlap I sent you a link to OneDrive. Please let me know if you cannot access or download the files. Thank you!

[pchlap]

Thanks for sharing these @hannahalme
I was able to reproduce the issue. It seems that we had assumed the RV would always extend above the LV inferiorly. This seems to not have happened in your case, perhaps just a differently oriented heart.

I have put in a fix here: #311
I'll wait for @rnfinnegan to approve this before merging. But feel free to test it out directly on that issue branch. Let me know what you think.

[hannahalme]

I tested it and it seems to work now. Thank you very much! :)

[EnriqueFV]

Hi, I wanted to try the solution of the branch but now I present problems on the dictionary that I passed to generate_left_ventricle_segments:

---------------------------------------------------------------------------
KeyError                                  Traceback (most recent call last)
Cell In[16], line 6
      1 m2 = {"LEFTVENTRICLE" : m['Ventricle_L'],
      2     "LEFTATRIUM"    : m['Atrium_L'],
      3     "RIGHTVENTRICLE": m['Ventricle_R'],
      4     "WHOLEHEART"    : m['Heart']}
----> 6 lv_segments = generate_left_ventricle_segments(contours=m2, verbose=False,
      7                                                 optimiser_tol_degrees=5,
      8                                                 optimiser_max_iter=16)

File /mnt/d/VHIR-IMM/CLAMP-PRECISED/.venv/lib/python3.11/site-packages/platipy/imaging/utils/ventricle.py:132, in generate_left_ventricle_segments(contours, label_left_ventricle, label_left_atrium, label_right_ventricle, label_heart, myocardium_thickness_mm, hole_fill_mm, optimiser_tol_degrees, optimiser_max_iter, min_area_mm2, verbose)
    129 label_mitral_valve = "MITRALVALVE"
    131 label_list = [label_left_ventricle, label_left_atrium, label_right_ventricle, label_heart]
--> 132 working_contours = copy.deepcopy({s: contours[s] for s in label_list})
    134 label_list.append(label_mitral_valve)
    136 output_contours = {}

File /mnt/d/VHIR-IMM/CLAMP-PRECISED/.venv/lib/python3.11/site-packages/platipy/imaging/utils/ventricle.py:132, in <dictcomp>(.0)
    129 label_mitral_valve = "MITRALVALVE"
    131 label_list = [label_left_ventricle, label_left_atrium, label_right_ventricle, label_heart]
--> 132 working_contours = copy.deepcopy({s: contours[s] for s in label_list})
    134 label_list.append(label_mitral_valve)
    136 output_contours = {}

KeyError: 'Ventricle_L'

Before, for the majority of the images worked when I construct the dictionary this way, but now I see that asks for the mitral valve. Is this something new?

[pchlap]

@EnriqueFV I don't think anything changed here. I haven't had a chance to dive deep back into this, but at first glance it seems like the look up key is wrong for the left ventricle in the dictionary passed in. Could it be that you need to pass in m instead of m2 to generate_left_ventricle_segments?