Demo: Generate a Point Cloud from a Single Patient Segmentation

Demo: Generate a Point Cloud from a Single Patient Segmentation#

This demo shows how to use the saxomode-pc CLI tool to generate epicardial and endocardial point clouds from .h5 segmentation files. We use a publicly available dataset available in https://www.ub.edu/mnms-2/.

Folder Structure#

project-root/
├── seg_files/
│   ├── 001.h5
│   └── 002.h5

Each .h5 file corresponds to one patient.

Required .h5 File Format#

Each .h5 file in seg_files/ must contain:

  • LV_mask: shape [slices, height, width] — binary left ventricle mask

  • RV_mask: shape [slices, height, width] — binary right ventricle mask

  • resolution: [px, py, pz] — voxel spacing in millimeters

Settings File#

A settings file is automatically generated in settings/. It contains point cloud fitting parameters:

{
  "mesh": {
    "fine": {
      "lax_smooth_level_epi": 80,
      "lax_smooth_level_endo": 50,
      "...": "..."
    }
  }
}

You may manually fine-tune these parameters:

Parameter

Description

lax_smooth_level_epi

Smoothness of epicardial LAX curves

lax_smooth_level_endo

Smoothness of endocardial LAX curves

Increasing these values produces smoother curves (useful for correcting motion or misalignment).

🚀 Run Point Cloud Generation#

To process a segmentation file for patient 001:

saxomode-pc --n 001

This will:

  • Load seg_files/001.h5

  • Generate a settings file settings/001.json (if not already present)

  • Output:

    • results/001/points_cloud_epi.csv

    • results/001/points_cloud_endo.csv

For each patient, a results folder is created (e.g., results/001/).
This folder contains several subfolders and files that help you check the quality of each processing step:

  • 01_Masks/: Visualizations of the original segmentation masks and extracted contours for each slice.

  • 02_ShaxBSpline/: Plots of the short-axis B-spline fits to the contours.

  • 03_LaxBSpline/: Plots of the long-axis B-spline fits.

  • 04_Contours/: Overlay plots of the generated contours with respect to the original segmentation stack—useful for visually checking the accuracy of the contour extraction and fitting.

  • 05_Point Cloud/: 3D visualizations of the generated point clouds.

You can open the images in these folders to visually inspect each step of the pipeline and ensure the results are as expected.

Check Fitting Accuracy#

To verify alignment with the original segmentation, generate a mesh:

saxomode-createmesh --n 001

This generates:

  • A .vtk mesh in: results/001/06_Mesh/

  • A .msh file

  • Diagnostic reports (fit quality, deviation from contours)

You can open the mesh in ParaView for inspection.

Contents