What should the default axon shape be: circle or ellipse(morphometrics) ?

As discussed in our previous meeting, we need to have consensus what the default axon shape should be for morphometrics computation . In order to determine that, I have tested the feature (#399) on the example test data as mentioned in RTD (see here).

The dataset consists of

• 1st SEM test image
  • image1_SEM
  • image2_SEM
• 2nd SEM test image
  • image1_SEM
• 1st TEM test image
  • image1_TEM

Morphometrics Analysis on these images

• 1st SEM test image (image1_SEM)

  - When axon shape is set as circle.

| gratio | axon_area | myelin_area | axon_diam | myelin_thickness | axonmyelin_area – | – | – | – | – | – | – 23 | 0.76019 | 38.318 | 27.9888 | 6.98484 | 1.101719 | 66.3068 41 | 0.50516 | 2.4108 | 7.0364 | 1.752006 | 0.858107 | 9.4472 44 | 0.694327 | 30.0713 | 32.3057 | 6.187727 | 1.362056 | 62.377 51 | 0.683596 | 14.1463 | 16.1259 | 4.244011 | 0.982176 | 30.2722 72 | 0.669229 | 4.6305 | 5.7085 | 2.428114 | 0.600056 | 10.339

image1_SEM_morphometrics_circle.xlsx

 - When axon shape is set as an ellipse

| gratio | axon_area | myelin_area | axon_diam | myelin_thickness | axonmyelin_area – | – | – | – | – | – | – 23 | 0.76019 | 38.318 | 27.9888 | 4.617765 | 0.971211 | 66.3068 41 | 0.50516 | 2.4108 | 7.0364 | 1.584517 | 0.842758 | 9.4472 44 | 0.694327 | 30.0713 | 32.3057 | 5.240084 | 1.384566 | 62.377 51 | 0.683596 | 14.1463 | 16.1259 | 3.620902 | 0.881783 | 30.2722 72 | 0.669229 | 4.6305 | 5.7085 | 1.542883 | 0.555395 | 10.339

image1_SEM_morphometrics_ellipse.xlsx

---

• 1st SEM test image (image2_SEM)

  - When axon shape is set as circle.

| gratio | axon_area | myelin_area | axon_diam | myelin_thickness | axonmyelin_area – | – | – | – | – | – | – 95 | 0.631449 | 21.1926 | 31.9579 | 5.194541 | 1.515919 | 53.1505 105 | 0.67661 | 3.0251 | 3.5828 | 1.962569 | 0.469012 | 6.6079 108 | 0.691071 | 44.8188 | 49.0269 | 7.554142 | 1.688458 | 93.8457 123 | 0.598044 | 10.9343 | 19.6378 | 3.731217 | 1.253911 | 30.5721 131 | 0.676415 | 39.2418 | 46.5257 | 7.068537 | 1.690733 | 85.7675

image2_SEM_morphometrics_circle.xlsx

  - When axon shape is set as an ellipse

| gratio | axon_area | myelin_area | axon_diam | myelin_thickness | axonmyelin_area – | – | – | – | – | – | – 95 | 0.631449 | 21.1926 | 31.9579 | 4.679203 | 1.456329 | 53.1505 105 | 0.67661 | 3.0251 | 3.5828 | 1.567248 | 0.469609 | 6.6079 108 | 0.691071 | 44.8188 | 49.0269 | 5.891441 | 1.549287 | 93.8457 123 | 0.598044 | 10.9343 | 19.6378 | 2.244691 | 1.245489 | 30.5721 131 | 0.676415 | 39.2418 | 46.5257 | 5.924539 | 1.623364 | 85.7675

image2_SEM_morphometrics_ellipse.xlsx

---

• 1st TEM test image (image1_TEM)

  - When axon shape is set as circle.

| gratio | axon_area | myelin_area | axon_diam | myelin_thickness | axonmyelin_area – | – | – | – | – | – | – 14 | 0.730813 | 0.286975 | 0.250344 | 0.604473 | 0.111326 | 0.537319 16 | 0.718671 | 0.2737 | 0.256225 | 0.590327 | 0.115544 | 0.529925 22 | 0.844741 | 1.3593 | 0.545581 | 1.315566 | 0.120897 | 1.904881 35 | 0.761378 | 0.1833 | 0.1329 | 0.483099 | 0.075704 | 0.3162 39 | 0.640557 | 0.1667 | 0.239575 | 0.460705 | 0.12926 | 0.406275

image1_TEM_morphometrics_circle.xlsx

     - When axon shape is set as an ellipse

| gratio | axon_area | myelin_area | axon_diam | myelin_thickness | axonmyelin_area – | – | – | – | – | – | – 14 | 0.730813 | 0.286975 | 0.250344 | 0.539991 | 0.106854 | 0.537319 16 | 0.718671 | 0.2737 | 0.256225 | 0.458162 | 0.107747 | 0.529925 22 | 0.844741 | 1.3593 | 0.545581 | 1.19527 | 0.119756 | 1.904881 35 | 0.761378 | 0.1833 | 0.1329 | 0.458095 | 0.083194 | 0.3162 39 | 0.640557 | 0.1667 | 0.239575 | 0.403173 | 0.098707 | 0.406275

image1_TEM_morphometrics_ellipse.xlsx

---

As you can clearly see that for the columns( axon_diam and myelin thickness), there is indeed a difference in the values when we select axon shape as circle or ellipse. But looking at these values, I am not sure how significant these differences are. So, I need @mathieuboudreau or @jcohenadad your insights on this.

[mariehbourget]

I disagree with the default being changed to ellipse. Now, people are using morphometrics with “circle” as the default. From a user point of view, it seems dangerous to suddenly change the default behavior, it could very easily lead to error while reporting results. In my opinion, documentation or warning can’t fully prevent those errors.

For example, if we go with “ellipse” as the default. Someone using the current notebook would have to change the call to get_axon_morphometrics to add the axon_shape=”circle”, otherwise, their results will change from previous computation.

What I would suggest is to keep the default as “circle”, which seems to be less used by our current users, but is still a valid approach as suggested by our litterature review on the topic.

I think it would also be essential to add information in the notebook and the code regarding the meaning of the shape for computation i.e. circle=equivalent diameter and ellipse=minor axis. Later, this information should also be included in RTD when morphometrics are added to CLI in #434.

[mathieuboudreau]

thank you @vs74 , these number look more reasonable now. I think I would change the default for the “ellipse” based approach, given that it seems this is what most ppl are using. What do you think @mathieuboudreau @mariehbourget ?

Yup, I agree with this. I’d like it if we could also add this information in some key locations throughout the project (e.g. README, RTD, Notebooks, and maybe even in some comments in the code) too so that people don’t report the wrong information in their publications after using the software.