Supervoxelizer

Note

This page is a reference documentation. It only explains the class signature, and not how to use it. Please refer to the Habitat Guide and Python API guide (v2.0) for usage.

class Supervoxelizer(*args, **kwargs)

Bases: Protocol

Partition one subject’s ROI into supervoxels.

Scientific motivation: voxel-level features are noisy, and clustering voxels directly across subjects tends to recover scanner scale differences rather than intratumoral heterogeneity. Aggregating within spatially coherent supervoxels, after per-subject normalisation, is what makes the subsequent population-level clustering biologically meaningful.

Implementations describe each region with the mean of the voxel features they partitioned, which is free once the partition exists and is what v0.1 always computed. That default is a summary, not a commitment: pass a SupervoxelFeatureExtractor to describe the same regions differently (per-supervoxel radiomics, for instance).

property spec: Spec

Return the algorithm specification.

__call__(field: VoxelFeatureField) → Supervoxelization

Group voxels into supervoxels and summarise them by feature mean.

Parameters:

field – Per-voxel features for one subject.

Returns:

The supervoxel partition together with per-supervoxel means.

__init__(*args, **kwargs)