Creating 3d Mesh from Z stack of ROIs

Question

I have a set of 3D ROI label stacks stored as TIFF files, where each ROI represents a single cell. I want to generate surface meshes(I use trimesh) suitable for downstream shape analysis.

I tried marching cubes:

from skimage import measure import trimesh import numpy as np mask = labels == lab_id # binary mask for one ROI verts, faces, normals, values = measure.marching_cubes(mask.astype(np.float32), level=0.5) mesh = trimesh.Trimesh(vertices=verts, faces=faces)

but Some meshes have missing surfaces, holes, or multiple disconnected “blobs. and Poisson surface reconstruction (Open3D) which didn't correctly capture the shapes.

I also tried

import trimesh coords = np.argwhere(mask) # N x 3 points pc = trimesh.points.PointCloud(coords) mesh = pc.convex_hull

But Convex hull loses all concavities, making every cell look roughly the same.

What is the recommended way to generate a nice and smooth closed-contour mesh from a 3D ROI stack, preserving concavities and avoiding blobs or missing surfaces?

Are there specific parameters in either of the methods I tried that help produce more reliable meshes?

Are you saying that the mesh this finds is not a correct representation of the mask, or are you saying that the mask variable contains voids / cavities that you want to fill? An example of the incorrect input / output would help me quite a bit. — Nick ODell
– Nick ODell, Commented Nov 14 at 21:14

Yigal Eilam · Accepted Answer · 2025-11-19 12:22:12Z

To triangulate a point cloud you can use Poisson surface reconstruction. It produces a smooth, watertight, closed-surface mesh that preserves concavities.

import numpy as np import open3d as o3d import trimesh # coords = np.argwhere(mask) # this is from your question: N x 3 point cloud # --- 1. Convert to Open3D point cloud --- pcd = o3d.geometry.PointCloud() pcd.points = o3d.utility.Vector3dVector(coords.astype(np.float32)) # --- 2. Estimate normals (required for Poisson) --- pcd.estimate_normals( search_param=o3d.geometry.KDTreeSearchParamHybrid( radius=4.0, # adjust based on cell size max_nn=60 ) ) pcd.orient_normals_consistent_tangent_plane(20) # --- 3. Poisson triangulation --- mesh, _ = o3d.geometry.TriangleMesh.create_from_point_cloud_poisson( pcd, depth=8 # depth 7–10 is typical; higher = more detail, slower ) # --- 4. Convert to trimesh --- tri = trimesh.Trimesh( vertices=np.asarray(mesh.vertices), faces=np.asarray(mesh.triangles) )

Collectives™ on Stack Overflow

Creating 3d Mesh from Z stack of ROIs

1 Answer 1

Comments

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Related