Problem with LongiSeg_plan_and_preprocess

[BaptBat]

Hi,

I'm trying to use LongiSeg with my dataset containing only 1 modality: T2w for multiples timepoints.
I have 2 questions / issues.

Firstly, does your implementation support non segmented data for context support ? Like, I have 1 timepoint, can I provide the previous timepoint without giving the corresponding previous segmentation ? I saw in a previous issue, you said you were supposed to do it ?

Secondly, here is my dataset structures:

-> dataset.json  patientsTr.json

-> imagesTr:
Borgne_ses07_bias_0000.nii.gz  Borgne_ses08_bias_0000.nii.gz

-> imagesTs:

-> labelsTr:
Borgne_ses07_bias.nii.gz  Borgne_ses08_bias.nii.gz

My dataset.json:

{
    "channel_names": {
        "0": "T2W"
    },
    "labels": {
        "background": 0,
        "CSF": 1,
        "WM": 2,
        "GM": 3,
        "Ventricle": 4
    },
    "numTraining": 2,
    "file_ending": ".nii.gz",
    "dataset_name": "Dataset001_Test"
}

And my patientsTr.json:

{
    "Borgne": [
        "Borgne_ses07_bias_0000.nii.gz",
        "Borgne_ses08_bias_0000.nii.gz"
    ]
}

When I run the command LongiSeg_plan_and_preprocess -d 1 --verify_dataset_integrity, I have the following error:

Fingerprint extraction...
Dataset001_Test
Using <class 'longiseg.imageio.simpleitk_reader_writer.SimpleITKIOWithReorient'> as reader/writer

####################
verify_dataset_integrity Done.
If you didn't see any error messages then your dataset is most likely OK!
####################

Using <class 'longiseg.imageio.simpleitk_reader_writer.SimpleITKIOWithReorient'> as reader/writer
100%|█████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:04<00:00,  4.21s/it]
multiprocessing.pool.RemoteTraceback:
"""
Traceback (most recent call last):
  File "/home/lbaptiste/miniconda3/envs/longiseg/lib/python3.10/multiprocessing/pool.py", line 125, in worker
    result = (True, func(*args, **kwds))
  File "/home/lbaptiste/miniconda3/envs/longiseg/lib/python3.10/multiprocessing/pool.py", line 51, in starmapstar
    return list(itertools.starmap(args[0], args[1]))
  File "/scratch/lbaptiste/Babofet_T2w/LongiSeg/longiseg/experiment_planning/dataset_fingerprint/fingerprint_extractor_longi.py", line 30, in analyze_patient
    image_files = dataset[s]['images']
KeyError: 'Borgne_ses07_bias_0000.nii.gz'
"""

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/home/lbaptiste/miniconda3/envs/longiseg/bin/LongiSeg_plan_and_preprocess", line 8, in <module>
    sys.exit(plan_and_preprocess_longi_entry())
  File "/scratch/lbaptiste/Babofet_T2w/LongiSeg/longiseg/experiment_planning/plan_and_preprocess_longi_entrypoints.py", line 180, in plan_and_preprocess_longi_entry
    extract_fingerprints(args.d, args.fpe, args.npfp, args.verify_dataset_integrity, args.clean, args.verbose)
  File "/scratch/lbaptiste/Babofet_T2w/LongiSeg/longiseg/experiment_planning/plan_and_preprocess_api.py", line 47, in extract_fingerprints
    extract_fingerprint_dataset(d, fingerprint_extractor_class, num_processes, check_dataset_integrity, clean,
  File "/scratch/lbaptiste/Babofet_T2w/LongiSeg/longiseg/experiment_planning/plan_and_preprocess_api.py", line 33, in extract_fingerprint_dataset
    return fpe.run(overwrite_existing=clean)
  File "/scratch/lbaptiste/Babofet_T2w/LongiSeg/longiseg/experiment_planning/dataset_fingerprint/fingerprint_extractor_longi.py", line 125, in run
    results = [i.get()[0] for i in r]
  File "/scratch/lbaptiste/Babofet_T2w/LongiSeg/longiseg/experiment_planning/dataset_fingerprint/fingerprint_extractor_longi.py", line 125, in <listcomp>
    results = [i.get()[0] for i in r]
  File "/home/lbaptiste/miniconda3/envs/longiseg/lib/python3.10/multiprocessing/pool.py", line 774, in get
    raise self._value
KeyError: 'Borgne_ses07_bias_0000.nii.gz'

I tried with "Borgne_ses07" (and also Borgne_ses08) in the patientsTr.json file, but got the same error.
My dataformat is well working with the basic UNet implementation: same filename as shown here for imagesTr and labelsTr, same dataset.json....

Thank you for your reply :)

[ykirchhoff]

Hi @BaptBat,

I unfortunately didn't find the time yet to imlpement having scans without segmentations. You can get around that by generating a pseudo segmentation only with the ignore label, which however won't give you any signal during training.
Your patientsTr.json should look something like this:
{
"Borgne": [
"Borgne_ses07_bias",
"Borgne_ses08_bias"
]
}
i.e. you strip the file ending and the modality identifier.

Best,
Yannick

[BaptBat]

Hi @ykirchhoff ,
Thanks ! I will check that and let you notice if I need help !

Best,
Baptiste