12 mhj/ 16-06-20041 Optimal Color Representation of Multi Spectral Data M.L.H. van Driel s462760 Supervisors: P. Sereda Prof. B.M. ter Haar Romeny.

Apr 18, 2023 1

Optimal Color Representation of Multi

Spectral Data

M.L.H. van Driel s462760Supervisors:P. SeredaProf. B.M. ter Haar Romeny

Apr 18, 2023 2

Contents• Introduction• Artherosclerotic plaques• Color Models

– RGB– HSV– CIE Lab

• Methods & Results• Conclusions• Discussion• Recommendations

Apr 18, 2023 3

Introduction

• From gray scale to multi spectral color images should improve the possibilities for tissue recognition and classification.

• Examples are multi spectral MRI measurements for artherosclerotic plaque classifications in medium and large arteries.

Apr 18, 2023 S.v.d.Ven (2004) 4

Artherosclerotic plaques

• Stable – Vulnerable• Important tissues

– Calcification– Fibrous tissue– Hemorrhage– Lipid– Lumen

Apr 18, 2023 5

Color Models

• RGB(Red Green Blue)

• HSV(Hue Saturation Value)

• CIE Lab(Commission Internationale d`Eclaraige

Lab)

Apr 18, 2023 http://www.bodoni.co.uk/colourworkshop.html

6

Color Models - RGB

• Primary colors• Range 0-255 (0-1)• Device dependant

Apr 18, 2023 http://www.ncsu.edu/scivis/lessons/colormodels/color_models2.html

7

Color Models - HSV

• Hue: color• Saturation:

dominance of hue• Value: lightness –

darkness• Range (0-1)

Apr 18, 2023 http://www.colourware.co.uk/cpfaq/q3-21.htm

8

Color Models – CIE Lab

• L – Lightness (0-100)• a – green red (-/+100)• b – blue yellow

(-/+100)• Device independent• Difference between 2

colors in the Lab space is an indication of the contrast

Apr 18, 2023 9

Methods & Results

• Input– Matching– Histogram Equalization

• The Optimal Color Model and Configuration

Apr 18, 2023 10

Methods & Results - Input

• 8 sets of 5 images with 3 tissues classified– T1 weighted (2D) TSE (1)– PD weighted TSE (2)– T1 weighted (3D) TFE (3)– Partial T2 weighted TSE (4)– T2 weighted TSE (5)

Apr 18, 2023 11

Methods & Results - input

• Matching– Same regions in different images should have the same

locations

• Histogram EqualizingHSV HSV HE

Apr 18, 2023 *http://www.cs.rit.edu/~ncs/color/t_convert.html

12

Methods & Results – The Optimal Color Model and

Configuration• Comparing the different Color Models

– Converting RGB and HSV to CIE Lab *

• Calculating the distances between the tissues

221

221

221 bbaaLL

Apr 18, 2023 13

Methods & Results – The Optimal Color Model and

Configuration• Red = CIE Lab• Green = RGB• Blue = HSV

20 40 60 80Mean

20

40

60

80

Minimal Distance

Apr 18, 2023 14

Methods & Results – The Optimal Color Model and

Configuration• Maximum of the minimal distance

– Cut off 30stepsize 30

Apr 18, 2023 15

Methods & Results – The Optimal Color Model and

ConfigurationGood configurations? the configurations

{{3,1,2},{3,2,1},{5,2,1},{5,1,2}} were found within the boundary conditions in 7 out of 8 cases (all within the Lab Color Model)

Lab 3 Lab 5

Lab 3 Lab 5

Apr 18, 2023 16

Conclusions

• To distinguish just arbitrary tissues CIE Lab is an appropriate Color Model

• In specific cases there can be better settings (in CIE Lab or even HSV) than in the arbitrary case

• RGB should not be considered an option

Apr 18, 2023 17

Discussion

• No Golden Truth• “Cut off 30” at least questionable• Histogram Equalization?• Other input

Apr 18, 2023 18

Discussion

• Histogram Equalization• Filters

0.2 0.4 0.6 0.8 1

0.2

0.4

0.6

0.8

1

Apr 18, 2023 19

Discussion – other inputs

LabToRGB RGB HSV

Apr 18, 2023 20

Recommendations

• 5D instead of 3D input• Differentiate for distinguishing

different tissues• Alternative for Histogram

Equalization• Other inputs

Apr 18, 2023 21

Thanks

• Petr Sereda• Bart ter Haar Romeny

• Woutjan Branderhorst• Martin Knýř