BPADBiomedical Physics with Applications to Disease

Intensity Segmentation Explorer

Segment a synthetic abdominal slice by intensity clustering and thresholding, with smoothing and K under your control.

segmentation
k-means
Otsu
smoothing

Preparing the simulation…

Original image

k-means segmentation (K=3, σ=0)

Intensity histogram with cluster centers

Cluster statistics

Cluster	Center (HU)	Voxels	% of image	Likely tissue
C1	-60.009	1807.000	44.1%	Air / fat
C2	28.987	1561.000	38.1%	Soft tissue
C3	120.271	728.000	17.8%	Enhancing tissue

The tumor spans multiple intensity classes

The enhancing rim, necrotic core and calcification of a real tumor occupy different HU ranges. K-means on intensity alone fragments a single lesion across clusters and merges tumor voxels with normal tissue that happens to share the same HU. Spatial smoothing adds coherence but blurs the true anatomical boundary — no setting of K or σ recovers "this whole region is the tumor" without spatial or supervised priors.