Automatic Diagnosis of Diabetic

Retinopathy
Using Retinal Fundus Image

MSc Research work by

IQBAL M.I, AIBINU A.M, GUBBAL N. and KHAN A.

Under the Supervision of
Mikael Nilsson
 The Project

-BTH, Optimum Biometric, in conjuction with Dr. Bergen Jack of Blekinge

Hospital intends applying Techniques of Digital Image Processing in the
Diagnosis of Diabetic Retinopathy (DR).

- Result of experiments shows that the system can be used to augment

Ophthalmologist work in the hospital.

- With further improvement, the system will totally replaced Screener’s

work in the Diagnosis of DR.
Content

‰Introduction

‰Methodology

‰Result Obtained

‰Graphical User Interface (GUI)

‰Conclusion and Recommendation

 Introduction

What is Diabetes ?

{ Diabetes is a disorder of metabolism Introduction

Diabetes ?
D.Retinopathy
Types of DR
Symptoms
{ Ability to Metabolise body Sugar impaired Diagnosing DR
Automating DR
Retina Images

{ Leads to build up of glucose in the blood Methodology

Pre-Processing
OFI
PFI
Comparison
{ Overflow into urine and passes out of the body Segmentation
SFI
Classifier
DCS
DCS Output
{ Secondary effects include Blindness, Stroke etc.
Result Obtained

GUI
{ More than 4% of Sweden population have Diabetes Conclusion
 Introduction

What is Diabetic Retinopathy?

Introduction
Diabetes ?
{ The effect of diabetes on the eye is called D.Retinopathy
Types of DR
Symptoms
Diagnosing DR
Diabetic Retinopathy Automating DR
Retina Images

Methodology
{ It damages small blood vessels of the Retina Pre-Processing
OFI
PFI
Comparison
Segmentation
{ Might lead to loss of vision SFI
Classifier
DCS
DCS Output
{ Early detection and treatment is recommended
Result Obtained

GUI
{ Treatment is by Laser Surgery Conclusion
 Introduction

Types of Diabetic Retinopathy (DR)

{Type 1 : Background Diabetic Retinopathy; the arteries Introduction

Diabetes ?
in the retina become weakened and leak, forming small, dot-like D.Retinopathy
Types of DR
haemorrhages Symptoms
Diagnosing DR
Automating DR
Retina Images
{Type 2 : Proliferate Diabetic Retinopathy; circulation
Methodology
problems cause areas of the retina to become oxygen-deprived or Pre-Processing
OFI
ischemic. New fragile, vessels develop as the circulatory system PFI
Comparison
attempts to maintain adequate oxygen levels within the retina Segmentation
SFI
Classifier
DCS
{Type 3 : Severe Diabetic Retinopathy; is continued DCS Output

abnormal vessel growth and scar tissue, which may cause Result Obtained

serious problems such as retinal detachment and glaucoma GUI

and gradual loss of vision Conclusion
 Introduction

Symptoms of DR

Introduction
{ Microaneurysms (Red Spots) Diabetes ?
D.Retinopathy

{ Haemorrhages (Bleeding) Types of DR

Symptoms
Diagnosing DR
{ Hard Exudates Automating DR
Retina Images

{ Soft Exudates Methodology

Pre-Processing
{ Vein-Artery Crossovers OFI
PFI
Comparison
{ Neovascularisation Segmentation
SFI
Classifier
DCS
DCS Output

All these artifacts are very small Result Obtained

GUI

Conclusion
 Introduction

Process of Diagnosing DR

Introduction
Diabetes ?
{ Involve capturing Retinal Image using specialised Cameras D.Retinopathy
Types of DR
Symptoms
Diagnosing DR
{ This Image is Observed and Screened by a Trained screener Automating DR
Retina Images

Methodology
{ If Diseases are detected, patient is referred to Specialised Ophthalmologist Pre-Processing
OFI
PFI
Comparison
Segmentation
{ Screening is tedious, time consuming and need specialised training SFI
Classifier
DCS
DCS Output
{ Hence need to augment the process
Result Obtained

GUI

Conclusion
 Introduction

Automating the Screener’s Work

The aim includes: Introduction

Diabetes ?
D.Retinopathy
Types of DR
Symptoms
Diagnosing DR
{ Developing a system that will be able to identify patients with Red Spot, Automating DR
Retina Images

Bleeding and Vein-Artery Crossover points. Methodology

Pre-Processing
OFI
{ Developing a MATLAB Based Graphic User Interface (GUI) for the display PFI
Comparison
Segmentation
SFI
of such Diseases. Classifier
DCS
DCS Output
{ Developing a MATLAB based GUI tool to be used by the Ophthalmologist
Result Obtained

GUI
in marking Fundus Images.
Conclusion
 Introduction

What does a Retina Image Look like ?

Introduction
Diabetes ?
D.Retinopathy
Types of DR
Symptoms
Diagnosing DR
Automating DR
Retina Images

Methodology
Pre-Processing
OFI
PFI
Comparison
Segmentation
SFI
Classifier
DCS
DCS Output

Result Obtained
Grey Level Image RGB Image GUI

Conclusion
 Methodology

Our Methodology !

Introduction
Diabetes ?
Retina Findus Image D.Retinopathy
Types of DR
Symptoms
Diagnosing DR
Automating DR
Retina Images
Image Pre Processing
Methodology
Pre-Processing
OFI
PFI
Image Segmentation Comparison
Segmentation
SFI
Classifier
DCS
Disease Classifier DCS Output

Result Obtained

GUI
Result: Normal / Abnormal
Conclusion
 Methodology

Image Pre-Processing Stage (IPS)

Introduction
Diabetes ?
Fundus D.Retinopathy
Types of DR
Image Symptoms
Diagnosing DR
Automating DR
Retina Images

Methodology
Color Space Conversion Pre-Processing
OFI
PFI
Comparison
Edge Zero Padding Segmentation
SFI
Classifier
DCS
DCS Output
Median Filtering
Result Obtained

Modified Adaptive Histogram Equalization GUI

Conclusion
 Methodology

IPS CONTD: ORIGINAL FUNDUS IMAGE(OFI)

z OFI is a grey level imageIntroduction

Diabetes ?
D.Retinopathy
Types of DR
Symptoms
z Intensity value from 0-255 Diagnosing DR
Automating DR
Retina Images

Methodology
z Badly illuminated Pre-Processing
OFI
PFI
Comparison
z Hardly visible regions Segmentation
SFI
Classifier
DCS
DCS Output
z Need to be enhanced Result Obtained
and equally illuminated GUI

Conclusion
 Methodology

IPS CONTD: PRE PROCESSED FUNDUS IMAGE(PFI)

z PFI is a grey level image

Introduction
Diabetes ?
D.Retinopathy
z Intensity value from 0-255 Types of DR
Symptoms
Diagnosing DR
z Well illuminated Automating DR
Retina Images

Methodology
z Visible regions and features Pre-Processing
OFI
PFI
Comparison
z Method used is Windowed Segmentation
Adaptive Histogram SFI
Classifier
Equalisation with Zero DCS
DCS Output
Filling
Result Obtained

GUI

Conclusion
 Methodology

COMPARING OFI WITH PFI

Introduction
Diabetes ?
D.Retinopathy
Types of DR
Symptoms
Diagnosing DR
Automating DR
Retina Images

Methodology
Pre-Processing
OFI
PFI
Comparison
Segmentation
SFI
Classifier
DCS
DCS Output

Result Obtained

GUI

„ OFI „ PFI Conclusion

 Methodology

Image Segmentation Stage (ISS)

Introduction
Pre-Processed Diabetes ?
Fundus Image D.Retinopathy
Types of DR
Symptoms
Diagnosing DR
Automating DR
Retina Images

Methodology
Class Segmentation using K-means Pre-Processing
Method OFI
PFI
Comparison
Segmentation
SFI
Classifier
DCS
DCS Output
Segmented
Result Obtained
Image
GUI

Conclusion
 Methodology

ISS CONTD: SEGMENTED FUNDUS IMAGE (SFI)

Introduction
Diabetes ?
z Well Segmented D.Retinopathy
Types of DR
Symptoms
Fundus Image using Diagnosing DR
Automating DR
K-Means Algorithm Retina Images

Methodology
Pre-Processing
OFI
z Segmented Image is PFI
Comparison
Segmentation
a Binary Image SFI
Classifier
DCS
DCS Output

Result Obtained

GUI

„ SFI Conclusion
 Methodology

Disease Classifier Stage (DCS)

Introduction
Diabetes ?
D.Retinopathy
Types of DR
Diseases of Interest are : Symptoms
Diagnosing DR
Automating DR
Retina Images

{ Microaneurysms (Red Spot) Methodology

Pre-Processing
OFI
PFI

{ Haemorrhages (Bleeding) Comparison

Segmentation
SFI
Classifier
DCS
{ Vein-Artery Crossover Points DCS Output

Result Obtained

GUI

Conclusion
 Methodology

Disease Classifier Stage (DCS)

Introduction
Diabetes ?
D.Retinopathy
Types of DR
Segmented Symptoms
Image Diagnosing DR
Automating DR
Retina Images

Methodology
Pre-Processing
OFI
Microaneurysm Detection PFI
Comparison
Segmentation
SFI
Classifier
Bleeding Detection DCS
DCS Output

Result Obtained
Crossover points Detection GUI

Conclusion
 Methodology

DCS CONTD: CLASSIFIER SECTION

Introduction
Diabetes ?
z CVision GUI D.Retinopathy
Types of DR
z Red Spots Symptoms
Diagnosing DR
Highlighted Automating DR
in Red Retina Images

z Bleeding Methodology
highlighted Pre-Processing
OFI
in blue PFI
Comparison
z Crossovers Segmentation
Highlighted SFI
Classifier
in Green DCS
DCS Output

Result Obtained

GUI

Conclusion
 Result Obtained

Final Decision

Introduction
Diabetes ?

Is the patient Normal or Abnormal ? D.Retinopathy

Types of DR
Symptoms

zSpecificity = 61% Diagnosing DR

Automating DR
Retina Images

zSensitivity = 98% Methodology

Pre-Processing
OFI
PFI
Comparison
Segmentation
Sensitivity is the percentage of abnormal fundus images SFI
Classifier
DCS
classified as abnormal DCS Output

Result Obtained
specificity can be defined as percentage of normal fundus GUI
images classify as normal Conclusion
 GUI

MATLAB BASED GUI- Other work Done

Introduction
Diabetes ?
D.Retinopathy
Types of DR
Symptoms
Diagnosing DR
Automating DR
Retina Images

Methodology
Pre-Processing
OFI
PFI
Comparison
Segmentation
SFI
Classifier
DCS
DCS Output

Result Obtained

z FIDA-GUI for Image Marking GUI

„ CVision GUI for DR Diagnosis
Conclusion
 Conclusion and Recommendation

Conclusion and Recommendations

Recommendations
Conclusion z Classifier section should be
z The system can assist in improved.
the Diagnosis of Diabetic z More diseases diagnosis
Retinopathy. should be added to the
system.
z Vein-Artery Crossover point
detection need
improvement.
z More Research resources
should be provided for this
work.
END
Automatic Diagnosis of
Diabetic Retinopathy