### DisTools examples: Brain MRI

Some exercises are defined on the basis of Brain MRI data. It is assumed that readers are familiar with PRTools and will consult the following pages where needed:

- PRTools User Guide, See at the bottom of the page for a TOC
- Introduction to DisTools
- Dissimilarity Representation Course
- The following packages should be in the Matlab path: PRTools, DisTools, PRDisData

This dataset is described in:

A. Ulas, R.P.W. Duin, U. Castellani, M. Loog, M. Bicego, V. Murino, M. Bellani, S. Cerruti, M. Tansella, and P. Brambilla, Dissimilarity-based Detection of Schizophrenia, *Proc. ICPR 2010 workshop on Pattern Recognition Challenges in FMRI Neuroimaging*, 2010.

It is based 13 dissimilarity measures between 14 ROIs in MRI brain images, resulting in 13×14 = 182 dissimilarity matrices. Here is an initial experiment.

`D = brainmri; % Loads a 13*14 cell array of dissimilarity matrices`

`E = zeros(size(D)); % Space for errors`

`F = zeros(size(D)); % Space for values of NEF`

`for j=1:13, for i=1:14`

`E(j,i)=`

`nne`

(D{j,i}); % LOO 1NN errors

`F(j,i)=nef(D{j,i}*makesym*pe_em); % NEF values`

`end, end`

To get an idea of the characteristics of the data we create a scatterplot of the results:

`A = prdataset([E(:) F(:)],repmat([1:13]',14,1));`

`scatterd(A);`

`xlabel Error`

`ylabel NEF`

`title 'NEF versus 1NN error for all BrainMRI dismats'`

`legend(num2str([1:13]'),'Location','EastOutside');`

The picture shows that the dissimilarity measures 1 and 7 are perhaps Euclidean. This can be verified by the names of the datasets in which the distance measures are reported:

`D{1,1}`

`D{7,1}`

#### Exercise

The above experiments are based on the given dissimilarities of individual dissimilarity matrices. Try to improve the results in the follwing ways:

- Classifiers in dissimilarity space.
- Classifiers in an embedded space.
- Classifiers based on a combination of dissimilarity measures: are there specific ROIs informative?
- Classifiers based on a combination of ROIs: what are the most inforamtive measures?
- Classifiers based on an overal combination of all dissimilarity matrices.

**elements:**
datasets
datafiles
cells and doubles
mappings
classifiers
mapping types.

**operations:**
datasets
datafiles
cells and doubles
mappings
classifiers
stacked
parallel
sequential
dyadic.

**user commands:**
datasets
representation
classifiers
evaluation
clustering
examples
support routines.

**introductory examples:**
Introduction
Scatterplots
Datasets
Datafiles
Mappings
Classifiers
Evaluation
Learning curves
Feature curves
Dimension reduction
Combining classifiers
Dissimilarities.

**advanced examples**.