Tumor Segmentation Techniques

Tsvetan Dimitrov, Fac. No.:201212015

• Image Segmentation

• MRI

• Segmentation Techniques

  • Thresholding

  • Region Growing

  • Clustering

  • Watershed Algorithm

  • Graph Based

Image Segmentation

• partitioning of a digital image into multiple sets of pixels (superpixels)

• enables easier analysis when presenting a simpler representation of the image

• regions with common characteristics depicted from same color, texture and intensity

MRI

• Magnetic Resonance Imaging

• magnetic field and pulses of radio wave energy to make pictures of organs and structures inside the body

• images produced by MRI are then processed using segmentation techniques

Segmentation Techniques

Thresholding / 1

• regions of the image belonging to different ranges of the gray scale

• each peak of the histogram of the image represents one region

Thresholding / 2

• the valley between the peaks represents a threshold value

• divide image into halves and compare their histograms to detect the tumor

Region Growing

• partitioning based on examining and adding of neighbouring pixels to a region

Clustering / 1

• organizing objects into groups based on some feature

• supervised and unsupervised

Clustering / 2

K-Means

• e.g. construct a representation of grey matter, white matter and tumor region

Clustering / 3

Fuzzy C-Means

Each data point belongs to a cluster to a degree specified by a membership value:

1. Divides a collection of N vectors into C fuzzy groups.

2. Finds a cluster centre in each group such that a cost function of dissimilarity measure is minimized.

Watershed Algorithm / 1

• watershed is defined by highpoints and ridgelines that descend into lower elevations and stream valleys

Watershed Algorithm / 2

• improves the primary results of K-Means in tumor segmentation

• applicable only if foreground and background of the image can be identified

Graph based

• nodes -> pixels

• edges -> neighbouring pixels

• use of dissimilarity metrics on common boundaries and pixels in each of the regions themselves 

Thank you!