Progress Report

May 2nd, 2006

General Overview

General Overview - Ctd.

Data with Noise: Initial Experiments

Data with Noise: Initial Experiments - Ctd.

Specificity Tested Under Noise

Choosing Data Noise Levels

Choosing Data Noise Levels - Ctd.

Example Gaussian Noise: Multiplied by 1

random_noise_level_1.png

Example Gaussian Noise: Multiplied by 2

random_noise_level_2.png

Example Gaussian Noise: Multiplied by 4

random_noise_level_4.png

Example Gaussian Noise: Multiplied by 8

random_noise_level_8.png

Example Gaussian Noise: Multiplied by 16

random_noise_level_16.png

Experimental Plan and Schedule

Experimental Plan and Schedule - Ctd.

Experimental Plan and Schedule - Ctd.

Assessment Experiments with Noise: An Early Glimpse

Experiment 1:
Noise 0:	 Gen: 2.58948 se: 0.0976118 Spec: 2.99433 se: 0.0177245
Noise 4:	 Gen: 2.7805 se: 0.0969311 Spec: 3.11209 se: 0.0166791
Noise 8:	 Gen: 3.22453 se: 0.107514 Spec: 3.57883 se: 0.0151328
Noise 16:	 Gen: 5.06954 se: 0.0981982 Spec: 5.60237 se: 0.0149359

Assessment Experiments with Noise: An Early Glimpse - Ctd.

Experiment 2:
Noise 0:	 Gen: 2.5967 se: 0.0984336 Spec: 3.0029 se: 0.0176939
Noise 4:	 Gen: 2.84286 se: 0.111833 Spec: 3.12515 se: 0.0162407
Noise 8:	 Gen: 3.22217 se: 0.100752 Spec: 3.5801 se: 0.015339
Noise 16:	 Gen: 5.12821 se: 0.101873 Spec: 5.63091 se: 0.014793

Noise in the Training Set (or Registered Set)

The Effect of Noise on Specificity

specificity-noise-effect-experiments.png
Based on 8 experiments, excludes error bars

The Effect of Noise on Generalisation

generalisation-ability-noise-effect-experiments.png
Based on 8 experiments, excludes error bars

Normality Test

The Next Set of Experiments

The Next Set of Experiments - Ctd.

Localised Registration Error

A Slice in Its Raw, Badly-scaled Form

slice_bad_contrast.png

A Properly-displayed Brain Slice

slice_recovered.png

An Accumulated (Shuffle) Difference Image

accumulated_difference_image_slice.png

Localised Registration Error - Ctd.

Average Shuffle Distance Images

average_shuffle_distance_image.png
Localised errors: essentially, the average of 8 shuffle difference images, built 'on the fly' and retained in memory

Practical Application: A Particular Gap

MIUA 2006 Paper Modified

Work on Entropy-based Measures