Surgical Planning

In this assignment, you are tasked with developing a segmentation algorithm. Your goal is to segment the pelvis, lumbar vertebrae, intervertebral discs, and the spinal cord within a CT dataset.

Objective: Implement a region growing algorithm to perform the segmentation, following the guidelines provided in lectures.

Setup

• Data Preparation: Transfer the pelvis_ct.nii.gz file from ILIAS to data/planning/pelvis_ct.nii.gz.

• Update Repository: Update your repository by running git pull --autostash.

• Implementation File: The segmentation.py file within assignments/planning/ directory contains a rough implementation. Complete the sections marked with ## TODO:.

  
  
  def region_grow(image, seed_point):
      """
      Performs a region growing on the image starting from 'seed_point'
      :param image: A 3D grayscale input image
      :param seed_point: The seed point for the algorithm
      :return: A 3D binary segmentation mask with the same dimensions as 'image'
      """
      segmentation_mask = np.zeros(image.shape, np.bool_)
      z, y, x = seed_point
      intensity = image[z, y, x]
      print(f'Image data at position ({x}, {y}, {z}) has value {intensity}')
      print('Computing region growing...', end='', flush=True)
  
      ## TODO: choose a lower and upper threshold
      threshold_lower = intensity
      threshold_upper = intensity
      _segmentation_mask = (np.greater(image, threshold_lower)
                            & np.less(image, threshold_upper)).astype(np.bool_)
  
      ## TODO: pre-process the segmented image with a morphological filter
  
      to_check = deque()
      to_check.append((z, y, x))
  
      while to_check:
          z, y, x = to_check.popleft()
  
          if _segmentation_mask[z, y, x]:
              # Mark the current point as visited
              _segmentation_mask[z, y, x] = False
              segmentation_mask[z, y, x] = True
  
              # These for loops will visit all the neighbors of a voxel and see if
              # they belong to the region
              for dz in range(-1, 2):
                  for dy in range(-1, 2):
                      for dx in range(-1, 2):
                          if dz == 0 and dy == 0 and dx == 0:
                              continue    # Skip the center point
                          nz, ny, nx = z + dz, y + dy, x + dx
  
                          ## TODO: implement the code which checks whether the current
                          ## voxel (nz, ny, nx) belongs to the region or not
  
                          ## OPTIONAL TODO: implement a stop criteria such that the algorithm
                          ## doesn't check voxels which are too far away
  
      # Post-process the image with a morphological filter
      structure = np.ones((3, 3, 3))
      segmentation_mask = ndimage.binary_closing(segmentation_mask, structure=structure).astype(np.bool_)
      
      print('\rComputing region growing... [DONE]', flush=True)
  
      return segmentation_mask
  

• Execution: Run your segmentation script via python cas/planning/planning.py to segment the specified structures. The segmentation does not need to be perfect, but the structures should be separated (see the 3D view above).

Testing your algorithm

Execute the planning.py script to test your algorithm. Use keyboard shortcuts to interact with the application, perform segmentation by clicking (several times), and save your work.

The script opens a window with a CT scan of a pelvis and spine model. The following keyboard shortcuts are available:

• 1-4 select active label. The segmented region (output of your algorithm) will then be labelled with the active label:

  0. None (default)

  1. Spinal cord

  2. Vertebrae

  3. Pelvis

  4. Discs

• click to start the region growing segmentation at the cursor position (seed point)

• space toggle segmentation mask overlay

• up or scroll move slice up

• down or scroll move slice down

• v visualise the resulting 3D surface models for the current segmentation mask

• r reset segmentation mask to zero

• s save segmentation mask to disk

• q quit (without saving)

Saving the output

• Save the completed segmentation to disk using the s key.

• Verify the saved segmentation mask by running the script show3d.py for a 3D visualisation: python cas/planning/show3d.py. The script will load and display the segmentation saved to disk.

• Create and save a screenshot of the 3D segmented models.

Code Submission

Submit a ZIP file named lastname_firstname_assignment1.zip on ILIAS containing:

1. The modified segmentation.py as lastname_firstname_assignment1_code.py.

2. Console output in a text file named lastname_firstname_assignment1_output.txt.

3. A screenshot of the 3D segmented model as lastname_firstname_assignment1_screenshot.png.

Online Questions

Complete the “Assignment 1 - Questions” on ILIAS:

• Answer all questions.

• Each question has only one correct answer.

• All questions are equally weighted. Incorrect answers will not result in point deductions.

• You are allowed only one attempt to complete the test.

Assignment Evaluation

This assignment constitutes 25% of your total assignment grade, split equally between:

• Code Evaluation (50%): points are awarded as follows:

  • 4 points for a working solution.

  • 3 points for only small errors.

  • 2 points for a substantial effort.

  • 1 point for substantial errors or minimal effort.

  • 0 points for no attempt or plagiarism.

• Online questions (50%)

Resources

• SciPy multidimensional image processing documentation