14.08.2013 Views

Docteur de l'université Automatic Segmentation and Shape Analysis ...

Docteur de l'université Automatic Segmentation and Shape Analysis ...

Docteur de l'université Automatic Segmentation and Shape Analysis ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

90 Chapter 4 Statistical shape mo<strong>de</strong>l of Hippocampus<br />

where W is the matrix which consists of all eigenvectors of XX T , <strong>and</strong> b is the<br />

vector <strong>de</strong>scribing the shape of X.<br />

The SSM built upon the optimized correspon<strong>de</strong>nce is evaluated in terms of its<br />

compactness, generalization ability <strong>and</strong> specificity (Davies et al., 2003).<br />

4.1.4.1 Compactness<br />

The compactness of the SSM is measured by the cumulative variance of the first<br />

M variation mo<strong>de</strong>s mo<strong>de</strong>s of mo<strong>de</strong>l<br />

Compactness(M) =<br />

∑Mm=1 λm<br />

∑n−1 m=1 λm<br />

, (4.52)<br />

where λm, m = 1, · · · , n − 1 are the <strong>de</strong>scendingly sorted eigenvalues of the covari-<br />

ance matrix Σ, with λ1 the largest eigenvalue.<br />

4.1.4.2 Generalization ability<br />

The generalization ability of the SSM measures the ability of the mo<strong>de</strong>l to represent<br />

an unseen shape. It is measured by a leave-one-out scheme<br />

Generalisability(M) = 1<br />

n<br />

n∑<br />

X<br />

i=1<br />

′ i(M) − Xi 2 , (4.53)<br />

where X ′ i(M) is the reconstruction of the shape Xi by the mo<strong>de</strong>l based on the<br />

training set excluding Xi with the first M mo<strong>de</strong>s, · is the L2-norm <strong>de</strong>fined on<br />

the vector space.<br />

4.1.4.3 Specificity<br />

The specificity is the expected distance between the shapes generated by mo<strong>de</strong>l<br />

<strong>and</strong> the shapes in the training set. It is evaluated by the generation shapes by the

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!