Datasets:

ash347
/

seedseg-demo

Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
case: string
synthetic: bool
n_markers: int64
markers: list<item: struct<centre_vox: list<item: double>, centre_ras_mm: list<item: double>, axis_vox: list< (... 103 chars omitted)
  child 0, item: struct<centre_vox: list<item: double>, centre_ras_mm: list<item: double>, axis_vox: list<item: doubl (... 91 chars omitted)
      child 0, centre_vox: list<item: double>
          child 0, item: double
      child 1, centre_ras_mm: list<item: double>
          child 0, item: double
      child 2, axis_vox: list<item: double>
          child 0, item: double
      child 3, angle_to_b0_deg: double
      child 4, delta_chi_ppm: double
      child 5, length_mm: double
      child 6, diameter_mm: double
distractors: list<item: struct<kind: string, centre_vox: list<item: double>, centre_ras_mm: list<item: double>, d (... 54 chars omitted)
  child 0, item: struct<kind: string, centre_vox: list<item: double>, centre_ras_mm: list<item: double>, delta_chi_pp (... 42 chars omitted)
      child 0, kind: string
      child 1, centre_vox: list<item: double>
          child 0, item: double
      child 2, centre_ras_mm: list<item: double>
          child 0, item: double
      child 3, delta_chi_ppm: double
      child 4, equivalent_diameter_mm: double
labels: struct<1: string, 2: string>
  child 0, 1: string
  child 1, 2: string
acquisition: struct<MagneticFieldStrength: double, RepetitionTime: double, EchoTime: double, FlipAngle: int64, Sl (... 150 chars omitted)
  child 0, MagneticFieldStrength: double
  child 1, RepetitionTime: double
  child 2, EchoTime: double
  child 3, FlipAngle: int64
  child 4, SliceThickness: double
  child 5, MRAcquisitionType: string
  child 6, ScanningSequence: string
  child 7, SequenceVariant: string
  child 8, PulseSequenceType: string
  child 9, effective_coils: int64
generator: struct<seed: int64, render: struct<shape_modes: list<item: double>>>
  child 0, seed: int64
  child 1, render: struct<shape_modes: list<item: double>>
      child 0, shape_modes: list<item: double>
          child 0, item: double
SequenceVariant: string
Synthetic: bool
PulseSequenceType: string
FlipAngle: int64
EchoTime: double
MRAcquisitionType: string
RepetitionTime: double
ImageType: list<item: string>
  child 0, item: string
SliceThickness: double
Description: string
MagneticFieldStrength: double
ScanningSequence: string
to
{'MagneticFieldStrength': Value('float64'), 'RepetitionTime': Value('float64'), 'EchoTime': Value('float64'), 'FlipAngle': Value('int64'), 'SliceThickness': Value('float64'), 'MRAcquisitionType': Value('string'), 'ScanningSequence': Value('string'), 'SequenceVariant': Value('string'), 'PulseSequenceType': Value('string'), 'Synthetic': Value('bool'), 'ImageType': List(Value('string')), 'Description': Value('string')}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 149, in get_rows_or_raise
                  return get_rows(
                      dataset=dataset,
                  ...<4 lines>...
                      column_names=column_names,
                  )
                File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 129, in get_rows
                  rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
                File "/src/services/worker/src/worker/utils.py", line 544, in safe_iter
                  yield from ds.decode(False) if ds.features else ds
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 3052, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2559, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2584, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 558, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 440, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 361, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2381, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2309, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              case: string
              synthetic: bool
              n_markers: int64
              markers: list<item: struct<centre_vox: list<item: double>, centre_ras_mm: list<item: double>, axis_vox: list< (... 103 chars omitted)
                child 0, item: struct<centre_vox: list<item: double>, centre_ras_mm: list<item: double>, axis_vox: list<item: doubl (... 91 chars omitted)
                    child 0, centre_vox: list<item: double>
                        child 0, item: double
                    child 1, centre_ras_mm: list<item: double>
                        child 0, item: double
                    child 2, axis_vox: list<item: double>
                        child 0, item: double
                    child 3, angle_to_b0_deg: double
                    child 4, delta_chi_ppm: double
                    child 5, length_mm: double
                    child 6, diameter_mm: double
              distractors: list<item: struct<kind: string, centre_vox: list<item: double>, centre_ras_mm: list<item: double>, d (... 54 chars omitted)
                child 0, item: struct<kind: string, centre_vox: list<item: double>, centre_ras_mm: list<item: double>, delta_chi_pp (... 42 chars omitted)
                    child 0, kind: string
                    child 1, centre_vox: list<item: double>
                        child 0, item: double
                    child 2, centre_ras_mm: list<item: double>
                        child 0, item: double
                    child 3, delta_chi_ppm: double
                    child 4, equivalent_diameter_mm: double
              labels: struct<1: string, 2: string>
                child 0, 1: string
                child 1, 2: string
              acquisition: struct<MagneticFieldStrength: double, RepetitionTime: double, EchoTime: double, FlipAngle: int64, Sl (... 150 chars omitted)
                child 0, MagneticFieldStrength: double
                child 1, RepetitionTime: double
                child 2, EchoTime: double
                child 3, FlipAngle: int64
                child 4, SliceThickness: double
                child 5, MRAcquisitionType: string
                child 6, ScanningSequence: string
                child 7, SequenceVariant: string
                child 8, PulseSequenceType: string
                child 9, effective_coils: int64
              generator: struct<seed: int64, render: struct<shape_modes: list<item: double>>>
                child 0, seed: int64
                child 1, render: struct<shape_modes: list<item: double>>
                    child 0, shape_modes: list<item: double>
                        child 0, item: double
              SequenceVariant: string
              Synthetic: bool
              PulseSequenceType: string
              FlipAngle: int64
              EchoTime: double
              MRAcquisitionType: string
              RepetitionTime: double
              ImageType: list<item: string>
                child 0, item: string
              SliceThickness: double
              Description: string
              MagneticFieldStrength: double
              ScanningSequence: string
              to
              {'MagneticFieldStrength': Value('float64'), 'RepetitionTime': Value('float64'), 'EchoTime': Value('float64'), 'FlipAngle': Value('int64'), 'SliceThickness': Value('float64'), 'MRAcquisitionType': Value('string'), 'ScanningSequence': Value('string'), 'SequenceVariant': Value('string'), 'PulseSequenceType': Value('string'), 'Synthetic': Value('bool'), 'ImageType': List(Value('string')), 'Description': Value('string')}
              because column names don't match

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

SeedSeg synthetic demo data

Synthetic T1-weighted prostate MRI with three gold fiducial markers each, for trying SeedSeg (web app, Python pipeline or OpenRecon container) without patient data. These images are synthetic. They are not patient scans.

Each case imitates the axial 2D FLASH acquisition SeedSeg was trained on (3 T, TR/TE 600/6.66 ms, flip angle 80°, 0.6875 × 0.6875 × 2 mm, 320 × 320 × 40). The images contain three gold markers and one to three simulated calcification-like or blood-product-like distractors in the prostate.

Files

File	Content
`demo-XXX_T1w.nii.gz`	T1w magnitude as from the scanner (with receive-coil shading; SeedSeg applies N4 itself)
`demo-XXX_T1w.json`	acquisition parameters
`demo-XXX_labels.nii.gz`	1 = gold marker (3 mm sphere at the marker centre), 2 = prostate
`demo-XXX_truth.json`	marker centres (voxel and RAS mm), axis, angle to B0, susceptibility; distractor positions and type
`dicom/demo-XXX/`	the same image as a DICOM MR series

Using the data

Web app: open https://astewartau.github.io/seedseg/ and click Load example data, or drop a *_T1w.nii.gz file or a dicom/demo-XXX folder.
Python: python scripts/deploy/seedseg_onnx.py --models <dir with seedseg-v2.json> --input demo-001_T1w.nii.gz --out results/demo-001
Compare the detections with demo-XXX_truth.json.

How the images were made

Anatomy. Organ contours from radiotherapy planning (body, bone, bladder, rectum, prostate, seminal vesicles, penile bulb, rectal spacer) of 112 prostate patients were combined into a probabilistic atlas with a model of shape variation. Each case is a new anatomy drawn from that population model.
Appearance. Tissue intensities, texture and noise use population statistics from the same acquisitions, with a multi-coil magnitude noise model, thin dark lines at tissue interfaces, bright inflow vessels and receive-coil shading.
Markers and distractors. Gold markers (1 × 3 mm cylinders) and distractors were inserted with a physics model: dipole field, through-slice dephasing, off-resonance slice displacement and the noise floor. The model was validated against 361 real marker voids.

No voxel of any patient image is in this dataset. Only label-derived population models and summary statistics were used. Each synthetic anatomy was compared with every source anatomy (mean Dice over bladder, rectum, prostate, bone marrow and seminal vesicles, aligned at the prostate). No case is closer to any source patient than the 95th percentile of how close the source patients are to each other.

Detection performance

These 8 cases were selected from 42 generated cases. In each of them the released SeedSeg ensemble finds all three markers and no other candidate. The web app's default setting (one network, no test-time mirroring) gives the same result on all 8.

Over all 42 generated cases, the released 5-fold ensemble found 111 of 126 markers, with 11 false positives among the top-3 detections. Almost every error came from a simulated distractor larger than 2 mm being ranked as a marker. In cross-validation on the 127 real patients, the same model configuration found 354 of 377 markers with 9 false positives. The simulated distractors are therefore harder than typical real calcifications.

Licence

CC BY 4.0. Please cite SeedSeg (see the repository) when using these data.

Downloads last month: 88

Total file size:

113 MB