Automatic measurement of ISNT and CDR on retinal images by means of - - PowerPoint PPT Presentation

▶
automatic measurement of isnt and cdr on retinal images
SMART_READER_LITE
LIVE PREVIEW

Automatic measurement of ISNT and CDR on retinal images by means of - - PowerPoint PPT Presentation

Sep 26, 2022 117 likes •554 views

Automatic measurement of ISNT and CDR on retinal images by means of a fast and efficient method based on mathematical morphology and active contours Rafael Verd-Monedero 1 , Juan Morales-Snchez 1 , Rafael Berenguer-Vidal 2 , Inmaculada

slide-1
SLIDE 1

Automatic measurement of ISNT and CDR on retinal images by means of a fast and efficient method based on mathematical morphology and active contours

Rafael Verdú-Monedero1, Juan Morales-Sánchez1, Rafael Berenguer-Vidal2, Inmaculada Sellés-Navarro3, Ana Palazón-Cabanes3

1Universidad Politécnica de Cartagena, Cartagena 30202, España 2Universidad Católica San Antonio, Murcia 30107, Spain 3Hospital Universitario Reina Sofía, Murcia 30003, Spain

8th International Work-Conference on the Interplay between Natural and Artificial Computation - IWINAC 2019

slide-2
SLIDE 2

Index

1

Introduction

2

Method Image acquisition and enhancement Automatic localization by morphological operations Accurate delineation by Parametric Active Contours

3

Results

4

Conclusions

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 2 / 25

slide-3
SLIDE 3

Index

1

Introduction

2

Method Image acquisition and enhancement Automatic localization by morphological operations Accurate delineation by Parametric Active Contours

3

Results

4

Conclusions

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 3 / 25

slide-4
SLIDE 4

Introduction

Spanish National projects - Instituto de Salud Carlos III

Subproject AES2017-PI17/00771 Extracción automática de características del nervio óptico de ambos ojos mediante procesado de imagen y reconocimiento de patrones para su integración en aplicación CAD de telemedicina Universidad Politécnica de Cartagena. Subproject AES2017-PI17-00821 Desarrollo y evaluación de una aplicación de telemedicina para el análisis anatómico conjunto del nervio óptico de ambos ojos Hospital Universitario Reina Sofía.

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 4 / 25

slide-5
SLIDE 5

Introduction

About Glaucoma

It is undoubtedly the most prevalent disease of optic neuropathies. This clinically silent ocular disease can lead to a progressive deterioration of the visual field causing eventually total blindness. Ocular anatomical alterations appear several years before the characteristic and irreversible lesions in the visual field, as a consequence of the death of the retina ganglion cells. Initial clinical explorations that may lead to suspected glaucoma

intraocular pressure levels, and the appearance of the optic nerve.

In these cases, other complementary test as the assessment of the retinal fibers layer and the study of the visual field should confirm the diagnosis.

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 5 / 25

slide-6
SLIDE 6

Introduction

About Glaucoma

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 6 / 25

slide-7
SLIDE 7

Introduction

About Glaucoma

Some anatomical parameters of the papilla can be used as reference indices in the early detection of glaucoma:

size and shape of the optic nerve (ISNT rule), the relationship between the size of the papillary excavation and the neuroretinal ring (cup-to-disc ratio, CDR), the configuration of the excavation depth, the exit position of the papillary vessels, the presence of peripapillary hemorrhages, the nerve fiber defect or the chorioretinal atrophy, ...

One of the drawbacks is the variability in size and appearance of the papilla in the normal population.

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 7 / 25

slide-8
SLIDE 8

Introduction

Reference indices based on anatomical parameters: ISNT rule

To differentiate normal from glaucomatous eyes in the clinical evaluation of the optic nerve head, normal eyes follow a characteristic configuration for the thickness of the disc rim: I ≥ S ≥ N ≥ T Eyes that deviate from the ISNT rule may need close monitoring for glaucoma.

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 8 / 25

slide-9
SLIDE 9

Introduction

Reference indices based on anatomical parameters: ISNT rule

The utility of this rule as a standalone criterion is not demonstrated for the early diagnosis of glaucoma. The predictability of this rule improves when it is complemented with other measurements and exploration data, such as, i.e., the cup-to-disc ratio. The presence of asymmetries in the papillary excavation between both eyes may be a suggestive factor of suffering from glaucoma.

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 8 / 25

slide-10
SLIDE 10

Introduction

Reference indices based on anatomical parameters: CDR

The Cup-to-disc ratio, CDR, compares the diameter of the cup portion with the total diameter of the

  • ptic disc.

Normal values of CDR are 0.3-0.4. The CDR is commonly used in

  • phthalmology as a measurement of

the evolution of glaucoma. As glaucoma advances, the cup enlarges until it occupies most of the disc area.

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 9 / 25

slide-11
SLIDE 11

Index

1

Introduction

2

Method Image acquisition and enhancement Automatic localization by morphological operations Accurate delineation by Parametric Active Contours

3

Results

4

Conclusions

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 10 / 25

slide-12
SLIDE 12

Method

Steps of the method

1

Image acquisition.

2

Preprocessing: Enhancement of the dynamic range and slight correction of the illumination.

3

Automatic localization of the optic disk and excavation by means

  • f morphological operations.

4

Accurate delineation of the optic disk and the excavation by two active contours.

5

Measurements: ISNT and CDR.

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 11 / 25

slide-13
SLIDE 13

Method

RGB channels of the color image

Original color image Red channel Green channel Blue channel

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 12 / 25

slide-14
SLIDE 14

Method

  • 1. Image acquisition

The fundus images were acquired by the Ophthalmology Service

  • f the Hospital General Universitario Reina Sofía (Murcia, Spain)

by means of a Topcon TRC-NW400 non-mydriatic retinal camera. These color images (RGB and 8 bit/channel) are in DICOM format, with a picture angle of 30◦ and an initial resolution of 1934 × 2576 pixels (downsampled to 600 × 800 pixels before processing).

  • 2. Image enhancement

Contrast stretching for a full widening of the dynamic range. Smooth local and adaptive histogram equalization for homogenizing the illumination of the images.

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 13 / 25

slide-15
SLIDE 15

Method

  • 3. a) Automatic location of optic disc by morphological operators

Red channel of the image, IR0. Closing and opening with a disk-shaped se R = 11: IR1 = ((IR0 • D11) ◦ D11) Top-hat using a R = 99 disk-shaped se: IR2 = (IR1 − (IR1 ◦ D99)) Threshold: MOD = thres(IR2)

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 14 / 25

slide-16
SLIDE 16

Method

  • 3. a) Automatic location of optic disc by morphological operators

Red channel of the image, IR0. Closing and opening with a disk-shaped se R = 11: IR1 = ((IR0 • D11) ◦ D11) Top-hat using a R = 99 disk-shaped se: IR2 = (IR1 − (IR1 ◦ D99)) Threshold: MOD = thres(IR2)

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 14 / 25

slide-17
SLIDE 17

Method

  • 3. a) Automatic location of optic disc by morphological operators

Red channel of the image, IR0. Closing and opening with a disk-shaped se R = 11: IR1 = ((IR0 • D11) ◦ D11) Top-hat using a R = 99 disk-shaped se: IR2 = (IR1 − (IR1 ◦ D99)) Threshold: MOD = thres(IR2)

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 14 / 25

slide-18
SLIDE 18

Method

  • 3. a) Automatic location of optic disc by morphological operators

Red channel of the image, IR0. Closing and opening with a disk-shaped se R = 11: IR1 = ((IR0 • D11) ◦ D11) Top-hat using a R = 99 disk-shaped se: IR2 = (IR1 − (IR1 ◦ D99)) Threshold: MOD = thres(IR2)

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 14 / 25

slide-19
SLIDE 19

Method

  • 3. a) Automatic location of optic disc by morphological operators

Red channel of the image, IR0. Closing and opening with a disk-shaped se R = 11: IR1 = ((IR0 • D11) ◦ D11) Top-hat using a R = 99 disk-shaped se: IR2 = (IR1 − (IR1 ◦ D99)) Threshold: MOD = thres(IR2)

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 14 / 25

slide-20
SLIDE 20

Method

  • 3. b) Automatic location of excavation by morphological operators

Blue channel of the image, IB0. ROI from the optic disc, IB1. Closing and opening IB2 = ((IB1 • D11) ◦ D11). Top-hat: IB3 = (IB2 − (IB2 ◦ D51)) Threshold: ME = thres(IB3)

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 15 / 25

slide-21
SLIDE 21

Method

  • 3. b) Automatic location of excavation by morphological operators

Blue channel of the image, IB0. ROI from the optic disc, IB1. Closing and opening IB2 = ((IB1 • D11) ◦ D11). Top-hat: IB3 = (IB2 − (IB2 ◦ D51)) Threshold: ME = thres(IB3)

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 15 / 25

slide-22
SLIDE 22

Method

  • 3. b) Automatic location of excavation by morphological operators

Blue channel of the image, IB0. ROI from the optic disc, IB1. Closing and opening IB2 = ((IB1 • D11) ◦ D11). Top-hat: IB3 = (IB2 − (IB2 ◦ D51)) Threshold: ME = thres(IB3)

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 15 / 25

slide-23
SLIDE 23

Method

  • 3. b) Automatic location of excavation by morphological operators

Blue channel of the image, IB0. ROI from the optic disc, IB1. Closing and opening IB2 = ((IB1 • D11) ◦ D11). Top-hat: IB3 = (IB2 − (IB2 ◦ D51)) Threshold: ME = thres(IB3)

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 15 / 25

slide-24
SLIDE 24

Method

  • 3. b) Automatic location of excavation by morphological operators

Blue channel of the image, IB0. ROI from the optic disc, IB1. Closing and opening IB2 = ((IB1 • D11) ◦ D11). Top-hat: IB3 = (IB2 − (IB2 ◦ D51)) Threshold: ME = thres(IB3)

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 15 / 25

slide-25
SLIDE 25

Method

  • 3. b) Automatic location of excavation by morphological operators

Blue channel of the image, IB0. ROI from the optic disc, IB1. Closing and opening IB2 = ((IB1 • D11) ◦ D11). Top-hat: IB3 = (IB2 − (IB2 ◦ D51)) Threshold: ME = thres(IB3)

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 15 / 25

slide-26
SLIDE 26
  • Method. Parametric Active Contours
  • 4. Accurate delineation by Parametric Active Contours

Coarse segmentation: MOD, ME. Least-squares ellipse fitting Fine segmentation with active contours

Green channel without veins. B-spline as shape function. External forces based on image gradient. Efficient implementation in the frequency domain

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 16 / 25

slide-27
SLIDE 27
  • Method. Parametric Active Contours
  • 4. Accurate delineation by Parametric Active Contours

Coarse segmentation: MOD, ME. Least-squares ellipse fitting Fine segmentation with active contours

Green channel without veins. B-spline as shape function. External forces based on image gradient. Efficient implementation in the frequency domain

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 16 / 25

slide-28
SLIDE 28
  • Method. Parametric Active Contours
  • 4. Accurate delineation by Parametric Active Contours

Coarse segmentation: MOD, ME. Least-squares ellipse fitting Fine segmentation with active contours

Green channel without veins. B-spline as shape function. External forces based on image gradient. Efficient implementation in the frequency domain

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 16 / 25

slide-29
SLIDE 29
  • Method. Parametric Active Contours
  • 4. Accurate delineation by Parametric Active Contours

Coarse segmentation: MOD, ME. Least-squares ellipse fitting Fine segmentation with active contours

Green channel without veins. B-spline as shape function. External forces based on image gradient. Efficient implementation in the frequency domain

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 16 / 25

slide-30
SLIDE 30
  • Method. Parametric Active Contours
  • 4. Accurate delineation by Parametric Active Contours

Coarse segmentation: MOD, ME. Least-squares ellipse fitting Fine segmentation with active contours

Green channel without veins. B-spline as shape function. External forces based on image gradient. Efficient implementation in the frequency domain

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 16 / 25

slide-31
SLIDE 31
  • Method. Parametric Active Contours
  • 4. Accurate delineation by Parametric Active Contours

Coarse segmentation: MOD, ME. Least-squares ellipse fitting Fine segmentation with active contours

Green channel without veins. B-spline as shape function. External forces based on image gradient. Efficient implementation in the frequency domain

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 16 / 25

slide-32
SLIDE 32
  • Method. Parametric Active Contours
  • 4. Accurate delineation by Parametric Active Contours

Coarse segmentation: MOD, ME. Least-squares ellipse fitting Fine segmentation with active contours

Green channel without veins. B-spline as shape function. External forces based on image gradient. Efficient implementation in the frequency domain

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 16 / 25

slide-33
SLIDE 33
  • Method. Parametric Active Contours
  • 4. Accurate delineation by Parametric Active Contours

Coarse segmentation: MOD, ME. Least-squares ellipse fitting Fine segmentation with active contours

Green channel without veins. B-spline as shape function. External forces based on image gradient. Efficient implementation in the frequency domain

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 16 / 25

slide-34
SLIDE 34
  • Method. Parametric Active Contours
  • 4. Accurate delineation by Parametric Active Contours

Coarse segmentation: MOD, ME. Least-squares ellipse fitting Fine segmentation with active contours

Green channel without veins. B-spline as shape function. External forces based on image gradient. Efficient implementation in the frequency domain

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 16 / 25

slide-35
SLIDE 35

Index

1

Introduction

2

Method Image acquisition and enhancement Automatic localization by morphological operations Accurate delineation by Parametric Active Contours

3

Results

4

Conclusions

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 17 / 25

slide-36
SLIDE 36

Results

Automatic measurements

Preliminary set of images. The main features considered in the early detection of glaucoma: ISNT and CDR. In the proposed approach, the cup-to-disc ratio has been computed from the active contours considering

the horizontal and vertical diameters that cross the center of mass

  • f both curves, CDRH and CDRV

the area of each curve, CDRS =

  • Scup

SOD

The measurements provided by the proposed method are complemented with the CDR provided by two expert

  • phthalmologists.

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 18 / 25

slide-37
SLIDE 37

Results

(a) Image #001 (b) Image #002

Figure: Results of ISNT and CDR computed with the proposed approach.

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 19 / 25

slide-38
SLIDE 38

Results

(a) Image #003 (b) Image #004

Figure: Results of ISNT and CDR computed with the proposed approach.

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 20 / 25

slide-39
SLIDE 39

Results

Table: CDR provided by two ophthalmologist and the proposed method.

Expert #1 Expert #2 Proposed method Im CDRH CDRV CDRH CDRV CDRH CDRV CDRS #1 0.4 0.5 0.4 0.4 0.45 0.48 0.47 #2 0.6 0.6 0.5 0.5 0.52 0.53 0.52 #3 0.4 0.4 0.4 0.3 0.45 0.46 0.45 #4 0.2 0.1 0.3 0.2 0.29 0.39 0.36

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 21 / 25

slide-40
SLIDE 40

Index

1

Introduction

2

Method Image acquisition and enhancement Automatic localization by morphological operations Accurate delineation by Parametric Active Contours

3

Results

4

Conclusions

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 22 / 25

slide-41
SLIDE 41

Conclusions

This paper has addressed a method to automatically locate and segment the optic disk and the excavation in retinal fundus images. This fast and efficient approach is based on mathematical morphology and a frequency-based implementation of active contours. Results show the accuracy of the proposed method to provide the CDR measurements. In ongoing research, the method will be compared with other approaches using databases such as STARE or ORIGA-light, whose performance is available.

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 23 / 25

slide-42
SLIDE 42

Thank you! Any question?

Slides available at http://www.tic.upct.es/rafael.verdu/publications.htm

  • r

rafael.verdu@upct.es

Verdú et al. (UPCT-UCAM-HGURS) IWINAC 2019 Thursday, 06.06.2019 24 / 25

slide-43
SLIDE 43

Automatic measurement of ISNT and CDR on retinal images by means of a fast and efficient method based on mathematical morphology and active contours

Rafael Verdú-Monedero1, Juan Morales-Sánchez1, Rafael Berenguer-Vidal2, Inmaculada Sellés-Navarro3, Ana Palazón-Cabanes3

1Universidad Politécnica de Cartagena, Cartagena 30202, España 2Universidad Católica San Antonio, Murcia 30107, Spain 3Hospital Universitario Reina Sofía, Murcia 30003, Spain

8th International Work-Conference on the Interplay between Natural and Artificial Computation - IWINAC 2019