[IMP] Duplicate Input DICOM Spatial Coordinates

[bluna301]

Is your enhancement request related to a problem? Please describe.
When running the MONAI CT TotalSegmentator MAP pipeline pythonically on an internal hospital study, the following error is produced by highdicom via the DICOMSegmentationWriterOperator:

(ct-totalsegmentator) bluna301@EW22-04661:~/ct-totalsegmentator-map$ ./scripts/model_run.sh
[info] [fragment.cpp:599] Loading extensions from configs...
[2025-04-02 09:44:27,202] [INFO] (root) - Parsed args: Namespace(log_level=None, input=PosixPath('/home/bluna301/ct-totalsegmentator-map/dicom_data/duplicate'), output=PosixPath('/home/bluna301/ct-totalsegmentator-map/output'), model=PosixPath('/home/bluna301/ct-totalsegmentator-map/model/model.ts'), workdir=None, argv=['my_app', '-i', '/home/bluna301/ct-totalsegmentator-map/dicom_data/duplicate', '-o', '/home/bluna301/ct-totalsegmentator-map/output', '-m', '/home/bluna301/ct-totalsegmentator-map/model/model.ts'])
[2025-04-02 09:44:27,202] [INFO] (root) - AppContext object: AppContext(input_path=/home/bluna301/ct-totalsegmentator-map/dicom_data/duplicate, output_path=/home/bluna301/ct-totalsegmentator-map/output, model_path=/home/bluna301/ct-totalsegmentator-map/model/model.ts, workdir=)
[2025-04-02 09:44:27,219] [INFO] (root) - End compose
[info] [gxf_executor.cpp:264] Creating context
[info] [gxf_executor.cpp:1797] creating input IOSpec named 'input_folder'
[info] [gxf_executor.cpp:1797] creating input IOSpec named 'dicom_study_list'
[info] [gxf_executor.cpp:1797] creating input IOSpec named 'study_selected_series_list'
[info] [gxf_executor.cpp:1797] creating input IOSpec named 'image'
[info] [gxf_executor.cpp:1797] creating input IOSpec named 'study_selected_series_list'
[info] [gxf_executor.cpp:1797] creating input IOSpec named 'text'
[info] [gxf_executor.cpp:1797] creating input IOSpec named 'output_folder'
[info] [gxf_executor.cpp:1797] creating input IOSpec named 'study_selected_series_list'
[info] [gxf_executor.cpp:1797] creating input IOSpec named 'seg_image'
[info] [gxf_executor.cpp:2208] Activating Graph...
[info] [gxf_executor.cpp:2238] Running Graph...
[info] [gxf_executor.cpp:2240] Waiting for completion...
[info] [greedy_scheduler.cpp:191] Scheduling 6 entities
[2025-04-02 09:44:27,347] [INFO] (monai.deploy.operators.dicom_data_loader_operator.DICOMDataLoaderOperator) - No or invalid input path from the optional input port: None
[2025-04-02 09:44:27,812] [INFO] (root) - Finding series for Selection named: Standard Axial CT Series
[2025-04-02 09:44:27,813] [INFO] (root) - Searching study, : 1.2.840.113619.2.405.3.185216268.497.1730719949.224
  # of series: 1
[2025-04-02 09:44:27,813] [INFO] (root) - Working on series, instance UID: 1.2.840.113619.2.405.3.185216268.497.1730719949.233
[2025-04-02 09:44:27,813] [INFO] (root) - On attribute: 'StudyDescription' to match value: '(.*?)'
[2025-04-02 09:44:27,813] [INFO] (root) -     Series attribute StudyDescription value: NM FDG PET CT SKULL TO MID THIGH AND CT N/C WITH CONTRAST
[2025-04-02 09:44:27,813] [INFO] (root) - Series attribute string value did not match. Try regEx.
[2025-04-02 09:44:27,813] [INFO] (root) - On attribute: 'Modality' to match value: '(?i)CT'
[2025-04-02 09:44:27,813] [INFO] (root) -     Series attribute Modality value: CT
[2025-04-02 09:44:27,813] [INFO] (root) - Series attribute string value did not match. Try regEx.
[2025-04-02 09:44:27,813] [INFO] (root) - On attribute: 'ImageType' to match value: ['PRIMARY', 'ORIGINAL', 'AXIAL']
[2025-04-02 09:44:27,813] [INFO] (root) -     Series attribute ImageType value: None
[2025-04-02 09:44:27,813] [INFO] (root) - On attribute: 'SliceThickness' to match value: [2, 5]
[2025-04-02 09:44:27,813] [INFO] (root) -     Series attribute SliceThickness value: None
[2025-04-02 09:44:27,813] [INFO] (root) - On attribute: 'SeriesDescription' to match value: '(?i)^(?!.*(cor|sag)).*$'
[2025-04-02 09:44:27,814] [INFO] (root) -     Series attribute SeriesDescription value: LDCT
[2025-04-02 09:44:27,814] [INFO] (root) - Series attribute string value did not match. Try regEx.
[2025-04-02 09:44:27,814] [INFO] (root) - Selected Series, UID: 1.2.840.113619.2.405.3.185216268.497.1730719949.233
[2025-04-02 09:44:27,814] [INFO] (root) - Series Selection finalized.
[2025-04-02 09:44:27,814] [INFO] (root) - Series Description of selected DICOM Series for inference: LDCT
[2025-04-02 09:44:27,814] [INFO] (root) - Series Instance UID of selected DICOM Series for inference: 1.2.840.113619.2.405.3.185216268.497.1730719949.233
[2025-04-02 09:44:35,110] [INFO] (ct_totalseg_operator.CTTotalSegOperator) - TorchScript model detected
[2025-04-02 09:44:35,111] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - Converted Image object metadata:
[2025-04-02 09:44:35,111] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - SeriesInstanceUID: 1.2.840.113619.2.405.3.185216268.497.1730719949.233, type <class 'str'>
[2025-04-02 09:44:35,111] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - SeriesDate: 20241111, type <class 'str'>
[2025-04-02 09:44:35,111] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - SeriesTime: 143622.000, type <class 'str'>
[2025-04-02 09:44:35,111] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - Modality: CT, type <class 'str'>
[2025-04-02 09:44:35,111] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - SeriesDescription: LDCT, type <class 'str'>
[2025-04-02 09:44:35,111] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - PatientPosition: FFS, type <class 'str'>
[2025-04-02 09:44:35,111] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - SeriesNumber: 2, type <class 'int'>
[2025-04-02 09:44:35,111] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - row_pixel_spacing: 0.976562, type <class 'float'>
[2025-04-02 09:44:35,111] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - col_pixel_spacing: 0.976562, type <class 'float'>
[2025-04-02 09:44:35,111] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - depth_pixel_spacing: 3.75, type <class 'float'>
[2025-04-02 09:44:35,111] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - row_direction_cosine: [1.0, 0.0, 0.0], type <class 'list'>
[2025-04-02 09:44:35,111] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - col_direction_cosine: [0.0, 1.0, 0.0], type <class 'list'>
[2025-04-02 09:44:35,111] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - depth_direction_cosine: [0.0, 0.0, 1.0], type <class 'list'>
[2025-04-02 09:44:35,111] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - dicom_affine_transform: [[ 9.76562000e-01  0.00000000e+00  0.00000000e+00 -2.50000000e+02]
 [ 0.00000000e+00  9.76562000e-01  0.00000000e+00 -2.50000000e+02]
 [ 0.00000000e+00  0.00000000e+00  3.72631833e+00 -1.32750000e+03]
 [ 0.00000000e+00  0.00000000e+00  0.00000000e+00  1.00000000e+00]], type <class 'numpy.ndarray'>
[2025-04-02 09:44:35,111] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - nifti_affine_transform: [[-9.76562000e-01 -0.00000000e+00 -0.00000000e+00  2.50000000e+02]
 [-0.00000000e+00 -9.76562000e-01 -0.00000000e+00  2.50000000e+02]
 [ 0.00000000e+00  0.00000000e+00  3.72631833e+00 -1.32750000e+03]
 [ 0.00000000e+00  0.00000000e+00  0.00000000e+00  1.00000000e+00]], type <class 'numpy.ndarray'>
[2025-04-02 09:44:35,111] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - StudyInstanceUID: 1.2.840.113619.2.405.3.185216268.497.1730719949.224, type <class 'str'>
[2025-04-02 09:44:35,111] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - StudyID: 5675, type <class 'str'>
[2025-04-02 09:44:35,111] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - StudyDate: 20241111, type <class 'str'>
[2025-04-02 09:44:35,112] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - StudyTime: 143203.000, type <class 'str'>
[2025-04-02 09:44:35,112] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - StudyDescription: NM FDG PET CT SKULL TO MID THIGH AND CT N/C WITH CONTRAST, type <class 'str'>
[2025-04-02 09:44:35,112] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - AccessionNumber: 11144480, type <class 'str'>
[2025-04-02 09:44:35,112] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - selection_name: Standard Axial CT Series, type <class 'str'>
[2025-04-02 09:44:39,779] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - Input of <class 'monai.data.meta_tensor.MetaTensor'> shape: torch.Size([1, 1, 334, 334, 774])
[2025-04-02 09:53:17,500] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - Post transform length/batch size of output: 1
[2025-04-02 09:53:17,827] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - Post transform pixel spacings for pred: tensor([0.9766, 0.9766, 3.7263], dtype=torch.float64)
[2025-04-02 09:53:17,828] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - Post transform pred of <class 'numpy.ndarray'> shape: (1, 512, 512, 312)
[2025-04-02 09:53:17,854] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - Output Seg image numpy array of type <class 'numpy.ndarray'> shape: (312, 512, 512)
[2025-04-02 09:53:17,861] [INFO] (monai_seg_inference_operator.MonaiSegInferenceOperator) - Output Seg image pixel max value: 104
[2025-04-02 09:53:18,488] [INFO] (root) - Finished writing DICOM instance to file /home/bluna301/ct-totalsegmentator-map/output/SR/1.2.826.0.1.3680043.8.498.47870571337236747815607421787958144492.dcm
[2025-04-02 09:53:18,496] [INFO] (monai.deploy.operators.dicom_text_sr_writer_operator.DICOMTextSRWriterOperator) - DICOM SOP instance saved in /home/bluna301/ct-totalsegmentator-map/output/SR/1.2.826.0.1.3680043.8.498.47870571337236747815607421787958144492.dcm
[error] [gxf_wrapper.cpp:100] Exception occurred for operator: 'dicom_seg_writer' - ValueError: Input image/frame positions are not unique according to the Dimension Index Pointers. The generated segmentation would be ambiguous. Ensure that source images/frames have distinct locations.

At:
  /home/bluna301/miniconda3/envs/ct-totalsegmentator/lib/python3.9/site-packages/highdicom/seg/content.py(668): get_index_values
  /home/bluna301/miniconda3/envs/ct-totalsegmentator/lib/python3.9/site-packages/highdicom/seg/sop.py(1829): __init__
  /home/bluna301/miniconda3/envs/ct-totalsegmentator/lib/python3.9/site-packages/monai/deploy/operators/dicom_seg_writer_operator.py(318): create_dicom_seg
  /home/bluna301/miniconda3/envs/ct-totalsegmentator/lib/python3.9/site-packages/monai/deploy/operators/dicom_seg_writer_operator.py(301): process_images
  /home/bluna301/miniconda3/envs/ct-totalsegmentator/lib/python3.9/site-packages/monai/deploy/operators/dicom_seg_writer_operator.py(280): compute

[error] [entity_executor.cpp:596] Failed to tick codelet dicom_seg_writer in entity: dicom_seg_writer code: GXF_FAILURE
[warning] [greedy_scheduler.cpp:243] Error while executing entity 41 named 'dicom_seg_writer': GXF_FAILURE
[info] [greedy_scheduler.cpp:401] Scheduler finished.
[error] [program.cpp:580] wait failed. Deactivating...
[error] [runtime.cpp:1649] Graph wait failed with error: GXF_FAILURE
[warning] [gxf_executor.cpp:2241] GXF call GxfGraphWait(context) in line 2241 of file /workspace/holoscan-sdk/src/core/executors/gxf/gxf_executor.cpp failed with 'GXF_FAILURE' (1)
[info] [gxf_executor.cpp:2251] Graph execution finished.
[error] [gxf_executor.cpp:2259] Graph execution error: GXF_FAILURE
Traceback (most recent call last):
  File "/home/bluna301/miniconda3/envs/ct-totalsegmentator/lib/python3.9/runpy.py", line 197, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/home/bluna301/miniconda3/envs/ct-totalsegmentator/lib/python3.9/runpy.py", line 87, in _run_code
    exec(code, run_globals)
  File "/home/bluna301/ct-totalsegmentator-map/my_app/__main__.py", line 25, in <module>
    CTTotalSegmentatorApp().run()
  File "/home/bluna301/ct-totalsegmentator-map/my_app/app.py", line 63, in run
    super().run(*args, **kwargs)
  File "/home/bluna301/miniconda3/envs/ct-totalsegmentator/lib/python3.9/site-packages/monai/deploy/operators/dicom_seg_writer_operator.py", line 280, in compute
    self.process_images(seg_image, study_selected_series_list, output_folder)
  File "/home/bluna301/miniconda3/envs/ct-totalsegmentator/lib/python3.9/site-packages/monai/deploy/operators/dicom_seg_writer_operator.py", line 301, in process_images
    self.create_dicom_seg(seg_image_numpy, dicom_series, output_dir)
  File "/home/bluna301/miniconda3/envs/ct-totalsegmentator/lib/python3.9/site-packages/monai/deploy/operators/dicom_seg_writer_operator.py", line 318, in create_dicom_seg
    seg = hd.seg.Segmentation(
  File "/home/bluna301/miniconda3/envs/ct-totalsegmentator/lib/python3.9/site-packages/highdicom/seg/sop.py", line 1829, in __init__
    self.DimensionIndexSequence.get_index_values(
  File "/home/bluna301/miniconda3/envs/ct-totalsegmentator/lib/python3.9/site-packages/highdicom/seg/content.py", line 668, in get_index_values
    raise ValueError(
ValueError: Input image/frame positions are not unique according to the Dimension Index Pointers. The generated segmentation would be ambiguous. Ensure that source images/frames have distinct locations.

Through parsing the DICOMSegmentationWriterOperator and the highdicom source code, it was determined that this error is caused by the input DICOM Series having multiple SOP instances with duplicate spatial coordinates (i.e. SliceLocation and ImagePositionPatient tags).

For this specific case, two SOP instances (Image #'s 156 and 157) have the exact same spatial coordinates:

• SliceLocation: -746.25
• ImagePositionPatient: [-250.000, -250.000, -746.250]

While they have the same spatial coordinates, the images do not appear visually identical (although they are very similar):

Image 156:

Image 157:

One explanation for the duplicate slices could be that this DICOM Series is a whole-body CT scan (312 total SOP instances) that has multiple AcquisitionNumber values (1-3) throughout the series (see below when DICOM Series is loaded in 3D Slicer). The duplicate slices occur at the most inferior slice of AcquisitionNumber = 2, and the most superior slice of AcquisitionNumber = 3.

Describe the solution you'd like
Ideally, there would be a way within MONAI Deploy App SDK to account for slices with duplicate spatial coordinates, perhaps by deleting one of the duplicate instances or by "skipping" that slice location and deleting both duplicates. Considering the DICOMSegmentationWriterOperator depends on highdicom, it may be easiest to employ a fix within the DICOMSeriesToVolumeOperator, perhaps within the prepare_series method where there is already existing functionality to remove slices.

Describe alternatives you've considered
Updating highdicom to account for duplicate input slices.

Additional context
Additional background on slices with duplicate spatial coordinates.

[MMelQin]

Thanks @bluna301 for also sharing the anonymized DICOM series. I loaded it up in a viewer and can concur with your initial assessment

The duplicates did happen where these changes took place,

• table speed lowered towards superior/head direction
• amperage and power increased towards the superior/head direction
• acquisition number changed from 2 to 3 (supine feet first, so higher number in the upper torso and head)

So, this is a multi-phased series, and in this case, looks to be in a continuous run except the one duplicate position.

The Series to volume operator does remove duplicates (will reconfirm), but the duplicates represented in the loaded series object is not removed. When the downstream Seg writer parses the original loaded series object and checking the seg image slices, discrepancies cause the runtime error.

So, here are my thoughts:

• assuming the change in image quality does not affect the inference
• enhance the DICOM loader to support the multi acquisition number use case and eliminate the one of the duplicate instance (with lower acquisition #?)
• the above will not work if multiple phases are interleaved, similar to the discussion on dcm2niix.
• So, should we just limit the loaded instances to be of the highest acquisition number?

[bluna301]

Thanks for reviewing @MMelQin. I agree with your suggested approach of rooting this out at the DICOM loading stage – great point about the slice removal at the DICOMSeriesToVolumeOperator stage not being sufficient to resolve the issue, as the error is caused by the source DICOM Series passed into the DICOMSegmentationWriterOperator.

I will look at the DICOMDataLoaderOperator to see if I can employ a fix.

[vikashg]

Thanks @MMelQin and @bluna301
@MMelQin I think you have more insight than I have on this. Do you think this is an edge case or is it a common issue ?

[MMelQin]

I wouldn't say it has been common, except for CTA (though often in the form of multi-frame instances, which we also don't support). It could become so though; dosage monitoring and control may contribute to it.
ChatGPT does think it is quite common (slightly mixing series with study; typically we'd have series without and series with contrasts).

• Q: Is that common to have multi-phase CT scan in that the same DICOM series would have multiple acquisition numbers among the instances with various scanner amperage etc?
• A: Yes, it is quite common to have multiple acquisition numbers or phases within the same DICOM series, especially in multi-phase CT scans. This typically occurs when a CT scan is performed over multiple phases (e.g., pre-contrast, arterial, venous, delayed phases), or when the scan parameters (like amperage, kVp, or scan timing) are altered between acquisitions to capture different aspects of the anatomy or pathology.

Here's why this can happen:

1. Multi-Phase Scans: For example, in a multi-phase CT angiogram, different phases of contrast injection might be captured to assess vascular structures at different times. Each phase would often have its own acquisition number, but all images would still be considered part of the same overall study, even though they might have different parameters such as amperage, kVp, or timing.

2. Variable Acquisition Parameters: In some cases, the scan parameters may be adjusted during the scan for specific reasons (e.g., to optimize contrast enhancement or minimize radiation exposure). This can lead to variations in the current (amperage) or other scan parameters for different segments of the scan. These acquisitions may still be grouped under the same study, but each might have its own acquisition number and set of DICOM instances.

3. Post-Processing: Modern CT systems may also allow for post-processing adjustments where the same raw data could be reprocessed to produce different views, reconstructions, or phases. These can appear in the DICOM series as distinct acquisitions but may represent different reconstructions of the same underlying data.

DICOM Structure:

• Study: Represents the entire examination.
• Series: Represents a set of images, typically acquired with the same parameters (e.g., the same contrast phase, slice thickness, etc.).
• Instance: A single image or slice.

In your case, the same DICOM series might have multiple acquisition numbers because the CT system assigns different numbers to acquisitions that were performed at different times or with different parameters, even though they are part of the same overall series. The series will be related but can have slightly different parameters such as amperage or slice thickness due to the scan protocol.

So, yes—this is relatively common, particularly in complex scans like those for vascular studies, oncology imaging, or cardiac CT, where multiple phases or protocols are necessary to evaluate different aspects of the anatomy.

[bluna301]

PR closed. Future work may need to be done to make this functionality optional (say, for example, the desired series for inference is an mDIXON MR series that has duplicate ImagePositionPatient values).