Detecting the excessive activation of the ciliaris muscle on thermal images Balázs Harangi Faculty of Informatics, University of Debrecen SSIP 2009, 10.

DDetecting the excessive activation etecting the excessive activation of the ciliaris muscle on thermal of the ciliaris muscle on thermal

imagesimages

Balázs HarangiFaculty of Informatics, University of Debrecen

SSIP 2009, 10 July 2009, Debrecen, Hungary

Overview

Background of the research

Steps of research

Program development

Conclusion and Discussion

Decision of ophthalmologist

23.04.18. 15:50

2

OverviewOverview

Aim of research:• Our primary aim in this field is to set up

a system which is able to alert, if the activity of the ciliaris muscle is suspected to be excessive.

• The main line of the research is to detect the extra quantity of heat caused by the excessive activity of the ciliaris muscle on thermal images.

• Final aim is to realize a system that is able to automatically diagnose.

Overview

Background of the research

Steps of research

Program development

Conclusion and Discussion

Decision of ophthalmologist

23.04.18. 15:50

3

Background of the researchBackground of the research

The ciliaris muscle

• A ring shaped muscle surrounds the crystalline lens in the eye.

• The muscle contracts when someone looks at a near object and relaxes when someone looks far.

Overview

Background of the research

Steps of research

Program development

Conclusion and Discussion

Decision of ophthalmologist

23.04.18. 15:50

4

Background of the researchBackground of the research

Problem of the ciliaris muscle

• The muscle does not relax in all cases, and thus, the crystalline lens does not flatten perfectly.

• The traditional ophthalmologic examination by a refractometer may provide a false dioptre value in this case.

• Fault of measurement cases vision improvement, head-ache, reading and other sight disorder.

Overview

Background of the research

Steps of research

Program development

Conclusion and Discussion

Decision of ophthalmologist

23.04.18. 15:50

5

Background of the researchBackground of the research

Aim and possible exploitation of the research

Since the ciliaris muscle is close to the exterior surface of the eye, we have the opportunity to take advantage of thermal monitoring of it.

Overview

Background of the research

Steps of research

Program development

Conclusion and Discussion

Decision of ophthalmologist

23.04.18. 15:50

6

Steps of ResearchSteps of Research

Images captured by Somatoinfra

384 x 288 pixels,8-bit intensity, 256-color images

Overview

Background of the research

Steps of research

Program development

Conclusion and Discussion

Decision of ophthalmologist

23.04.18. 15:50

7

Steps of ResearchSteps of Research

The usage of grayscale images

Usually color palettes are applied for displaying in order to let smaller differences to be easily detectable for a human observer, but for simplicity, we change the color representation to grayscale.

Overview

Background of the research

Steps of research

Program development

Conclusion and Discussion

Decision of ophthalmologist

23.04.18. 15:50

8

Steps of ResearchSteps of Research

Image normalization

)(

)(

D

EThe temperature of the skin depends on the external weather or the internal temperature of the examine room. For these reasons, we inserted a normalization step into our system to eliminate these differences.

Overview

Background of the research

Steps of research

Program development

Conclusion and Discussion

Decision of ophthalmologist

23.04.18. 15:50

9

Steps of ResearchSteps of Research

Localization of the eye

We modeled the eyes with ellipses which are subdivided into subregions.Thus, we can focus to the interesting regions only.

Overview

Background of the research

Steps of research

Program development

Conclusion and Discussion

Decision of ophthalmologist

23.04.18. 15:50

10

Steps of ResearchSteps of Research

First-order statistical descriptors:

WALK RUN

• Mean of intensity histogram:

• Variance of intensity histogram:

• Skewness of intensity histogram:

• Kurtosis of intensity histogram :

• Energy of intensity histogram :

• Entropy of intensity histogram :

)(1

0

iipL

iH

)()( 21

0

ipiH

L

iH

3

31

0

)()(

H

H

L

i

ipi

3)()(

4

41

0

H

H

L

i

ipi

)(21

0

ipL

i

)())((log2

1

0

ipipL

i

Overview

Background of the research

Steps of research

Program development

Conclusion and Discussion

Decision of ophthalmologist

23.04.18. 15:50

11

Steps of ResearchSteps of Research

Training and classification

We gain 144 dimensional feature vectors per eye if all the subregions are involved. After then we considered the kNN classifier (with k=10) to decide whether a test image was labeled as healthy or diseased.

Overview

Background of the research

Steps of research

Program development

Conclusion and Discussion

Decision of ophthalmologist

23.04.18. 15:50

12

Program developmentProgram development

Program language: The sourcecode is written in Matlab, we used the

following tools:

– Image Processing Toolbox:• imread(); imshow()• imellipse(); vertex(); poly2mask()• graycomatrix(); graycoprops()

– Statistics Toolbox:• skewness(); kurtosis()

– and Bioinformatics Toolbox:• knnclassify()

Overview

Background of the research

Steps of research

Program development

Conclusion and Discussion

Decision of ophthalmologist

23.04.18. 15:50

13

Functions:• Normalization of heat scale• Definition of region of eyes

• Extraction of features• Classification

Program developmentProgram development

Overview

Background of the research

Steps of research

Program development

Conclusion and Discussion

Decision of ophthalmologist

23.04.18. 15:50

14

Conclusion and DiscussionConclusion and Discussion

DatabaseOur initial training database contains

20 healthy and 20 diseased images

manually labeled by a clinical expert.

(diseased) (healty)

Overview

Background of the research

Steps of research

Program development

Conclusion and Discussion

Decision of ophthalmologist

23.04.18. 15:50

15

Conclusion and DiscussionConclusion and Discussion

Result:Our simple algorithm is tested on small databased. Result of this test:

FP FN Overall

Our method (selected regions) 3/20 2/20 75%

Our method (all regions) 4/20 3/20 65%

Overview

Background of the research

Steps of research

Program development

Conclusion and Discussion

Decision of ophthalmologist

23.04.18. 15:50

16

Conclusion and DiscussionConclusion and Discussion

Fault results:

True positive False positive

Overview

Background of the research

Steps of research

Program development

Conclusion and Discussion

Decision of ophthalmologist

23.04.18. 15:50

17

Conclusion and DiscussionConclusion and Discussion

Fault results:

True negative False negative

Overview

Background of the research

Steps of research

Program development

Conclusion and Discussion

Decision of ophthalmologist

23.04.18. 15:50

18

Conclusion and DiscussionConclusion and Discussion

Summary• Basic functions of the finally system are ready and

acting.• Results of the decision are good enough, but we

have to refine.

Plans• A current database should be extended • More specify results• Automatic location of the eye region• Finding the critical point of eye on normal picture

(corner, pupil, ciliaris muscle)• Find an appropriate physical model to get rid of

thermal distortion of orbit.

Overview

Background of the research

Steps of research

Program development

Conclusion and Discussion

Decision of ophthalmologist

23.04.18. 15:50

19

Decision of Decision of ophthalmologistophthalmologist

One way to more exact results:The practical diagnostics is based on comparing the thermal value of the ciliaris muscle with the centre of the cornea.

Overview

Background of the research

Steps of research

Program development

Conclusion and Discussion

Decision of ophthalmologist

23.04.18. 15:50

20

Thank you!Thank you!

Thank you for your attention!