Adding model to segment bones #134

gattia · 2022-12-06T01:34:22Z

Starting a pull request while working on this to get feedback. A few things are done differently than other models.

Currently loading model + weights from .h5 file (vs. building model & loading weights)
- This requires storing shape of original images used during training, reshaping inputs to that shape, segmenting, and then returning the shape. https://github.com/gattia/DOSMA/blob/adcebe3df0556eaddac95b3d7501d313f361a367/dosma/models/stanford_qdess_bone.py#L180-L215
- Added option to "resample_images" which if True uses this strategy of using the original model (and resampling shape). If False, presumably we will re-build the original model and load the weights. This is essentially a placeholder for me to get code to re-generate the model(s) directly - I was worried at first I had something in the network that wasn't fully convolutional (Dense layer), but I think we are good. Right now this throws an error (will implement in future).
Use argmax to create a single segmentation of each tissue (instead of thresholding each tissue separately).
- https://github.com/gattia/DOSMA/blob/adcebe3df0556eaddac95b3d7501d313f361a367/dosma/models/stanford_qdess_bone.py#L199
Added option connected_only to model.generate_mask this is because the bones produced a few spurious points in tests. This is on by default for all tissues.
- https://github.com/gattia/DOSMA/blob/adcebe3df0556eaddac95b3d7501d313f361a367/dosma/models/stanford_qdess_bone.py#L217
The model segments medial/lateral tissues separately.
- To keep convention with other models, it then creates a "men" and "tc" tissue that just combines med/lat segments.
  - https://github.com/gattia/DOSMA/blob/adcebe3df0556eaddac95b3d7501d313f361a367/dosma/models/stanford_qdess_bone.py#L165-L176
- Also saved the "all" segmentation mask that is a single mask with all tissues (not one-hot/binary).
  - https://github.com/gattia/DOSMA/blob/adcebe3df0556eaddac95b3d7501d313f361a367/dosma/models/stanford_qdess_bone.py#L152-L154

The main things I see myself wanting to do/add are:

Create function to build original model so we can skip resampling images up/down?
Add a model using same architecture as other stanford_qdess.py models - I think the segmentation performance on this model vs. that are pretty much the same.

- qdess bone seg returns each individual tissue - Added map of combined maps.

…_seg

ad12 · 2023-08-14T00:16:05Z

@gattia thanks for the PR! could you lint your files and provide some same screenshots of the segmentations

gattia · 2023-08-29T16:20:59Z

Looks like it's getting hungup during CI because Python 3.6 isn't available anymore?

Performed lint on my side + added a photo of the current segs. I'd love to add the two remaining things I have listed on my PR comments, but they're not necessary for this PR and maybe I can do at a later date.

- if pd > 1.2 then use the new .close syntax. If older, then use old .save syntax. https://stackoverflow.com/questions/76090979/xlsxwriter-object-has-no-attribute-save-did-you-mean-save

- using non-tuple for multidimensional indexing was deprecated. - also remove debug print statements

- if scipy >= 2.0.0 then import center_of_mass from scipy.ndimage - else, if earlier then import from scipy.ndimage.measurements.

- This is only available in python>3.10, therefore removing the requirement in flake8

- Dont need docs for each specific version...

- Number was due to an error in a specifc version (that numba is now past) - SITK was due to change in basis (don't accept non-orthonormal) - but this hasnt been an issue running tests upto now.

- np.complex is deprecated.

- Different versions of tensorflow/keras were returning `tuple` or `tf.TensorShape`. - Create new function to handle both - refactor utils to create this function and the current model loading util functions.

- Codecoverage was throwing rate limiting errors and thus failing CI. -https://github.com/ad12/DOSMA/actions/runs/10459944994/job/28965177349#step:7:83 - Therefore, add if statement to no push codecov under these circumstances.

The current implementation worked somewhat to fix errors with specific image acquisitions, but was somewhat specific to sagittal images. It has now been updated to reformat and then export images to sitk as would be expected for a specific image acquisition type (e.g., sagittal, coronal, axial). Essentially, based on if its sag, ax, or coral, the `mv` will be reformatted. Then, this reformatted image along with the affine/spacing information a new sitk Image will be created. Finally, if the dicom was not collected in this standard orientation, the user can still choose to flip the volume along any of the three axes using boolean flags. New tests were also written to ensure the reformatting is such that the image is roughly what it would be like if SITK itself loaded it.

… available.

This is based on a simplification of equations 6/7 of the original Sveinsson et al. paper "A Simple Analytic Method for Estimating T2 in the Knee from DESS". This is required because the new Mensa product sequence does not store these header fields.

gattia and others added 24 commits December 1, 2022 17:31

Adding qdess bone segmentation model

af3da65

Bone seg return all tissues

fb4d584

- qdess bone seg returns each individual tissue - Added map of combined maps.

Import bone model to models sub-module

b925b39

Add bone seg model to models submodeuls namespace

9e7bf99

Done super SegModel because not same funcs

d71179e

Dont compile loaded bone seg model

695a575

fix tuple concatenation syntax

373d206

expand dims for feature layer - use features first

5c1e475

Adding qdess bone segmentation model

dfbdb0b

Bone seg return all tissues

cbd039a

- qdess bone seg returns each individual tissue - Added map of combined maps.

Import bone model to models sub-module

cda4896

Add bone seg model to models submodeuls namespace

6309433

Done super SegModel because not same funcs

2011737

Dont compile loaded bone seg model

dd92a61

fix tuple concatenation syntax

6dd76cc

expand dims for feature layer - use features first

95793ec

Merge branch 'bone_seg' of https://github.com/gattia/DOSMA into bone_seg

288b65b

Add comments - update postprocess reshaping

3989833

Merge remote-tracking branch 'refs/remotes/origin/bone_seg' into bone…

e6221ae

…_seg

Option (Default) get largest connected components

ba97176

Update docs, provide option for re-building model

b368fb5

Updating documentation

adcebe3

Apply linting

672eb78

autoformatting code with make

ee10a10

gattia added 2 commits August 29, 2023 08:18

Lint / autoformat

0950d45

Add bone segs figures

4db4709

gattia and others added 2 commits October 17, 2023 08:37

stanford_qdess_bone to return uint8 masks

1c13dd1

update StanfordQDessBoneINet2D to be flexible on model image size

69c0229

gattia added 30 commits August 16, 2024 16:29

update np.int to int

2ea5c44

Add medvolume debugging print statements

84cf578

Get rid of strict=True for zip. so compatible with python<3.10

dedc05a

Updating excel writer syntax to use .close vs .save

ef0783c

- if pd > 1.2 then use the new .close syntax. If older, then use old .save syntax. https://stackoverflow.com/questions/76090979/xlsxwriter-object-has-no-attribute-save-did-you-mean-save

Update h5py loading data syntax

b9cf7cf

keep kers version logger - in case we go back to keras in future

732b44a

turn np index into tuple

e64a0db

- using non-tuple for multidimensional indexing was deprecated. - also remove debug print statements

update readme re: unit test environment variables.

7e27a16

Updating np.bool to be bool b/c np.bool deprecated

0e85754

Fix deprecated scipy.ndimage.measurements

c36d518

- if scipy >= 2.0.0 then import center_of_mass from scipy.ndimage - else, if earlier then import from scipy.ndimage.measurements.

Update install instructions to install from source.

7988b87

Update gitignore to not log nipype logs.

f2f6907

ignore zip() strict=False in flake8 errors

046e77f

- This is only available in python>3.10, therefore removing the requirement in flake8

Linting updates.

f352308

Fixing sni import errors.

ae5fcd8

Drop CI testing for python 3.6 and add 3.9/3.10

f2b505a

Drop CI testing for python 3.6 and add 3.9/3.10

d3f5279

Drop CI Documentation for 3.7

34f1e1e

- Dont need docs for each specific version...

Remove pinned numba & simpleitk from CI

2cec41f

- Number was due to an error in a specifc version (that numba is now past) - SITK was due to change in basis (don't accept non-orthonormal) - but this hasnt been an issue running tests upto now.

update np.complex to use builtin complex

055f282

- np.complex is deprecated.

Flexibly get tensor shapes

327f37e

- Different versions of tensorflow/keras were returning `tuple` or `tf.TensorShape`. - Create new function to handle both - refactor utils to create this function and the current model loading util functions.

Update docs to specify we are not python 3.7+

c15a8a4

linting

4b5f3a8

No codecov for forked repositroy pull request updates

97bbb31

- Codecoverage was throwing rate limiting errors and thus failing CI. -https://github.com/ad12/DOSMA/actions/runs/10459944994/job/28965177349#step:7:83 - Therefore, add if statement to no push codecov under these circumstances.

Linting

ac65e62

Update new sitk tests to use dicom data - thus not to run if data not…

7596c2d

… available.

Merge branch 'bone_seg' of https://github.com/gattia/DOSMA into bone_seg

2f2b960

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Adding model to segment bones #134

Adding model to segment bones #134

gattia commented Dec 6, 2022

ad12 commented Aug 14, 2023 •

edited

Loading

gattia commented Aug 29, 2023

Adding model to segment bones #134

Are you sure you want to change the base?

Adding model to segment bones #134

Conversation

gattia commented Dec 6, 2022

ad12 commented Aug 14, 2023 • edited Loading

gattia commented Aug 29, 2023

ad12 commented Aug 14, 2023 •

edited

Loading