Structural Changes -Color Images

1
Structural Changes -Color Images Algorithms for Structural and Functional Change Analysis from Multi-Modal Retinal Fundus Images Harihar Narasimha-Iyer 1 , James M. Beach 2 , Ali Can 3 , Badrinath Roysam 1 , Charles V. Stewart 1 , Howard Tanenbaum 4 1 Rensselaer Polytechnic Institute, Troy, NY-12180 2 Institute for Technology Development, Stennis Space Center, Mississippi 39529, USA 3 Woods Hole Oceanographic Institution, Woods Hole, Massachusetts 02543, USA. 4 The Center for Sight, 349 Northern Blvd., Albany, New York 12204, USA. This work was supported in part by CenSSIS, the Center for Subsurface Sensing and Imaging Systems, under the Engineering Research Centers Program of the National Science Foundation (Award Number EEC- 9986821) Type of Color Type of Color Change Change Significance Significance Increase in Red Appearance of bleeding or microaneurysm Decrease in Red Disappearance of bleeding or microaneurysm Increase in Yellow Appearance of exudate Decrease in Yellow Disappearance of exudate Types of Changes in Non- vascular Regions 1. M. J. Cree, J.A. Olson, K.C. McHardy, J.V. Forrester and P.F. Sharp, “Automated microaneurysm detection,” IEEE Int. Conf. on Image Processing, vol. 3, pp.699-702, Lausanne, Switzerland, 1996. 2. M.J. Cree, J.A. Olson, K.C. McHardy, P.F. Sharp, J.V. Forrester, “A fully automated comparative microaneurysm digital detection system,” Eye, vol. 11, pp. 622-628,1998. 3. K.A. Goatman, M. J. Cree, J.A. Olson, J.V. Forrester and P.F. Sharp, “Automated measurement of microaneurysm turnover,” Investigative ophthalmology and Visual Science, vol.44,5335-5341, 2003. 4. Z.B. Sbeh, L.D. Cohen, G. Mimoun and G. Coscas, “A new approach of geodesic reconstruction for drusen segmentation in eye fundus images,” IEEE Trans. Medical. 5. J.M. Beach, K.J. Schwenzer, S. Srinivas, D. Kim, and J.S. Tiedeman, “Oximetry of retinal vessels by dual-wavelength imaging: calibration and influence of pigmentation,” Journal of Appl. Physiology. vol. 86,748-758, 1999. 6. H. Narasimha-Iyer, J. M. Beach, B. Khoobehi, J. Ning, H. Kawano and B. Roysam, “Algorithms for Automated Oximetry along the Retinal Vascular Tree from Dual- Wavelength Fundus Images,” Accepted for publication, Journal of Biomedical Optics, May 2005. 7. H. Narasimha-Iyer, A. Can, B. Roysam, C.V. Stewart, H.L. Tanenbaum, A. Majerovics and H. Singh, " Robust Detection and Classification of Longitudinal Changes in Color Retinal Fundus Images for Monitoring Diabetic Retinopathy," Accepted for publication in IEEE Transactions on Biomedical Engineering, September 2005. State of the art Only a few methods have been described for quantifying the dynamic nature of diabetic retinopathy from a time series of retinal images. Cree et al. [1, 2] detect microaneurysms from a region of interest around the fovea from images at distinct time points and compare them to find changes. Studies of microaneurysm turnover were also made by Goatman et al. [3]. They detected microaneurysms from baseline and follow-up angiograms, registered the images and categorized the microaneurysms into three classes namely, static, new and regressed. Sbeh and Cohen [4] segment drusen based on a morphological method called geodesic reconstruction and study the evolution of drusen over time..All these methods have the limitation that they handle only a certain kind of lesion and also the changes are obtained from individual segmentation of the images. The present work contributes multiple advances over the above literature, overcoming many of the noted limitations. The net result is an algorithm and a systematic non- limiting framework that allows a broad range of longitudinal changes to be detected and classified with a high degree of reliability. The Framework Vascular Linking 1 2 3 (,, )/ (,, ); (,, ) (,, ); (,, ) (,, ) 2; i green i red i green j green i green j green f R xy R xy f R xy R xy f R xy R xy Bayesian Change Classifier ( | ) ( | ), m =1,2...5,k m. k m PC X PC X for Features used for the classification: Bayesian classification rule: Bayesian model selection used to associate descriptions with regions of changes in the vasculature. Information from the tracing algorithm (IUS) and the change classifier are combined to get the likelihood functions. Describing Vascular Changes 1 2 ( , ) i region i segment XOR IUS IUS 1 1 , , 2 2 ( , , /region ihasdisappeared) ( /region ihasdisappeared) ( , /region ihasdisappeared) im jn im jn pII pII p 1 2 1 2 ( , /region ihasdisappeared) ( , / ) x x x region pI I pI I red down ( , /region ihasdisappeared)= 2 im jn p confidence factorassociated w ith the IU S outputs. Changes in Nonvascular regions Changes in Nonvascular regions t i Modality m Modality k Image Understanding System IUS Assisted Image Correction Image Understanding System IUS Assisted Image Correction Registration Application Specific Pre- Application Specific Pre- Processing Processing Application Specific Pre- Application Specific Pre- Processing Processing Spectral Change Features Structural/Functional Change Analyzer ij T i IU S j IU S i C j C i D j D ij SP i IU S j IU S (Objects) (Corrected Images) (Features for change analysis) (Corrected & Processed Images) High level change descriptions t j Modality m Modality k Identification of Vessel Type Retinal Image Understanding System Functional Changes-Dual Wavelength Images Changes in Blood Oxygen Saturation Room Air Breathing Pure O 2 Breathing OD Measurement Day 1 Trace Mask –IUS 1 Day 2 Trace Mask – IUS 2 Vascular Changes Change Regions Absorption spectra of HbO 2 ( ) and Hb ( ) , 570 , ( ) log out seg im in seg I OD I 1 2 3 ; ( ) ( ); ( ) ( ). im jn im jn f ODR f ODR f ODR ODR 600 570 ( ) ( ) . ( ) im im im OD ODR OD Retinal Vessel Oximetry The image at 570nm is traced and the trace segments are linked and named according to the hierarchy Image correction involves median filtering to remove any noise Measure the optical density (OD) based on the minimum reflectance inside the vessel and the average outside reflectance

description

Absorption spectra of HbO 2 ( ) and Hb ( ). t i. t j. Modality k. Modality k. Modality m. Modality m. Image Understanding System. Image Understanding System. Registration. (Objects). IUS Assisted Image Correction. IUS Assisted Image Correction. (Corrected Images). - PowerPoint PPT Presentation

Transcript of Structural Changes -Color Images

Page 1: Structural Changes -Color Images

Structural Changes -Color Images

Algorithms for Structural and Functional Change Analysis from Multi-Modal Retinal Fundus ImagesHarihar Narasimha-Iyer1, James M. Beach2 , Ali Can3, Badrinath Roysam1, Charles V. Stewart1, Howard Tanenbaum4

1Rensselaer Polytechnic Institute, Troy, NY-121802Institute for Technology Development, Stennis Space Center, Mississippi 39529, USA

3Woods Hole Oceanographic Institution, Woods Hole, Massachusetts 02543, USA. 4The Center for Sight, 349 Northern Blvd., Albany, New York 12204, USA.

This work was supported in part by CenSSIS, the Center for Subsurface Sensing and Imaging Systems, under the Engineering Research Centers Program of the National Science Foundation (Award Number EEC-9986821)

Type of Color ChangeType of Color Change SignificanceSignificance

Increase in RedAppearance of bleeding or

microaneurysm

Decrease in RedDisappearance of bleeding or

microaneurysm

Increase in Yellow Appearance of exudate

Decrease in Yellow Disappearance of exudate

Types of Changes in Non-vascular Regions

1. M. J. Cree, J.A. Olson, K.C. McHardy, J.V. Forrester and P.F. Sharp, “Automated microaneurysm detection,” IEEE Int. Conf. on Image Processing, vol. 3, pp.699-702, Lausanne, Switzerland, 1996.

2. M.J. Cree, J.A. Olson, K.C. McHardy, P.F. Sharp, J.V. Forrester, “A fully automated comparative microaneurysm digital detection system,” Eye, vol. 11, pp. 622-628,1998.

3. K.A. Goatman, M. J. Cree, J.A. Olson, J.V. Forrester and P.F. Sharp, “Automated measurement of microaneurysm turnover,” Investigative ophthalmology and Visual Science, vol.44,5335-5341, 2003.

4. Z.B. Sbeh, L.D. Cohen, G. Mimoun and G. Coscas, “A new approach of geodesic reconstruction for drusen segmentation in eye fundus images,” IEEE Trans. Medical. Imaging, vol. 20, no.12, pp.1321-1333, 2001.

5. J.M. Beach, K.J. Schwenzer, S. Srinivas, D. Kim, and J.S. Tiedeman, “Oximetry of retinal vessels by dual-wavelength imaging: calibration and influence of pigmentation,” Journal of Appl. Physiology. vol. 86,748-758, 1999.

6. H. Narasimha-Iyer, J. M. Beach, B. Khoobehi, J. Ning, H. Kawano and B. Roysam, “Algorithms for Automated Oximetry along the Retinal Vascular Tree from Dual-Wavelength Fundus Images,” Accepted for publication, Journal of Biomedical Optics, May 2005.

7. H. Narasimha-Iyer, A. Can, B. Roysam, C.V. Stewart, H.L. Tanenbaum, A. Majerovics and H. Singh, " Robust Detection and Classification of Longitudinal Changes in Color Retinal Fundus Images for Monitoring Diabetic Retinopathy," Accepted for publication in IEEE Transactions on Biomedical Engineering, September 2005.

State of the artOnly a few methods have been described for quantifying the dynamic nature of diabetic retinopathy from a time series of retinal images. Cree et al. [1, 2] detect microaneurysms from a region of interest around the fovea from images at distinct time points and compare them to find changes. Studies of microaneurysm turnover were also made by Goatman et al. [3]. They detected microaneurysms from baseline and follow-up angiograms, registered the images and categorized the microaneurysms into three classes namely, static, new and regressed. Sbeh and Cohen [4] segment drusen based on a morphological method called geodesic reconstruction and study the evolution of drusen over time..All these methods have the limitation that they handle only a certain kind of lesion and also the changes are obtained from individual segmentation of the images. The present work contributes multiple advances over the above literature, overcoming many of the noted limitations. The net result is an algorithm and a systematic non-limiting framework that allows a broad range of longitudinal changes to be detected and classified with a high degree of reliability.

The Framework

Vascular Linking

1

2

3

( , , ) / ( , , );

( , , ) ( , , );

( , , ) ( , , ) 2;

i green i red

i green j green

i green j green

f R x y R x y

f R x y R x y

f R x y R x y

Bayesian Change Classifier

( | ) ( | ), m=1,2...5,k m.k mP C X P C X for

Features used for the classification:

Bayesian classification rule:

Bayesian model selection used to associate descriptions with regions of changes in the vasculature. Information from the tracing algorithm (IUS) and the change classifier

are combined to get the likelihood functions.

Describing Vascular Changes

1 2 ( , )iregion i segment XOR IUS IUS

1 1, ,2 2( , , / region i has disappeared) ( / region i has disappeared)

( , / region i has disappeared)

im jn

im jn

p I I p I I

p

1 2 1 2( , / region i has disappeared) ( , / )x xx region

p I I p I I red down

( , / region i has disappeared ) = 2im jnp

confidence factor associated with the IUS outputs.

Changes in Nonvascular regions

Changes in Nonvascular regions

ti

Modality m Modality k

Image Understanding System

IUS Assisted Image Correction

Image Understanding System

IUS Assisted Image Correction

Registration

Application Specific Pre-Application Specific Pre-Processing Processing

Application Specific Pre-Application Specific Pre-ProcessingProcessing

Spectral Change Features

Structural/Functional Change Analyzer

ijT

iIUS jIUS

iCjC

iD jD

ijSP

iIUS jIUS

(Objects)

(Corrected Images)

(Features for change analysis)

(Corrected & Processed Images)

High level change descriptions

tj

Modality m Modality k

Identification of Vessel Type

Retinal Image Understanding System

Functional Changes-Dual Wavelength Images

Changes in Blood Oxygen Saturation

Room Air Breathing Pure O2 BreathingOD Measurement

Day 1

Trace Mask –IUS1

Day 2

Trace Mask – IUS2

Vascular Changes

Change Regions

Absorption spectra of HbO2 ( ) and Hb ( )

,570

,

( ) log out segim

in seg

IOD

I

1

2

3

;( )

( );

( ) ( ).

im

jn

im jn

f ODR

f ODR

f ODR ODR

600

570

( )( ) .

( )im

imim

ODODR

OD

Retinal Vessel Oximetry

The image at 570nm is traced and the trace segments are linked and named according to the hierarchy

Image correction involves median filtering to remove any noise

Measure the optical density (OD) based on the minimum reflectance inside the vessel and the average

outside reflectance