Gurgaon Sector 90 Call Girls ( 9873940964 ) Book Hot And Sexy Girls In A Few ...
Enhancements to a Computer : Assisted Screening Technology for Diabetic Retinopathy by Sheila John
1. 1
Enhancements to a computer-
assisted screening technology
for diabetic retinopathy: system
redesign based on our pilot
study in indian setting
2. 2
Authors
Sheila John, Kulasekaran S, Supriti M,
Keerthi Ram, Mohanasankar S, Rajiv Raman,
Badrinath S.S
Sankara Nethralaya
Healthcare Technology Innovation Centre, IIT
Madras
3. 3
Diabetic Retinopathy
(DR) in India
More than 60 million diabetic people in India
Prevalence of DR is 18% in diabetic population
Significant prevalence in both rural and urban population
Acute shortfall of ophthalmologists
1 per 100,000 population
Need: Preventive eye-care through early
identification
5. 5
Screening technology for DR
Existing computer-assisted DR screening solutions
Europe:
UK: iGrading, Portugal: Retmarker
Americas:
US: IDx-DR, Canada: CARA
State of the art performance: sensitivity 97% at 47%
specificity †
† Retinal imaging and image analysis, Abramoff et
al, IEEE rev. Biomed. Engg, 2010
6. 6
Screening technology for DR
India Academic research activities at IIT-KGP,
IIIT-Hyd, IIT-Madras, few Engg. Colleges
DR screening research activities world-over
more than 200 peer-reviewed publications since
2003
7. 7
IITM DR screening system
Normal anatomy detection Clinical signs detection
Red lesion
detectionOptic disc and
macula detection
Blood vessel
segmentation
Bright lesion
detection
Small red dots
detection
Image gradabilityInput
image
Analytics
DR Referral decision
Grading system based
on International
Clinical Diabetic
retinopathy Disease
Severity Scale (ICDR)
5 severity levels
Normal anatomy detection Clinical signs detection
Red lesion
detectionOptic disc and
macula detection
Blood vessel
segmentation
Bright lesion
detection
Small red dots
detection
Image gradabilityInput
image
Analytics
DR Referral decision
Grading system based
on International
Clinical Diabetic
retinopathy Disease
Severity Scale (ICDR)
5 severity levels
8. 8
IIT Madras DR screening
system
Consists of modules for detecting disease signs,
and analytics for providing a referable vs non-
referable decision
Developed and benchmarked using 2000 publicly
available fundus images acquired in clinical
settings
Refinements to algorithms for working in Indian
settings : 85.9% sensitivity at 83% specificity
9. 9
Pilot retrospective study
and observations
Observed performance on subset of 200 images
of SN-DREAMS Retrospective study
Mydriatic, 45 degree Retinal images
Includes images with media opacity, severe
pathology, and lower quality of image capture, for
observing performance
Grading by ophthalmologist following ICDR
guideline – 5 severity levels
10. 10
Pilot retrospective study
and observations
Needs to handle image gradability and non-
mydriatic imaging
Separate analytics for diabetic macular edema
and Proliferative diabetic retinopathy
Designed to find new cases of DR, but also Laser
treated cases.
Evaluation of inter-observer variability and
consensus should be carried out
11. 11
Module for image gradability
Image
preprocessing
Enhanced image
Structure
distribution
Colour
distribution
Contrast Illumination SNRHomogeneity Moments
Quality prediction
Gradability score
Reference images for
good gradability
Reference images for
poor gradability
Quality parameters
Retinal image
Evaluated on 240 images: 82% sensitivity at 80% specificity
12. 12
Redesign: module for CSME
Includes module for accurate localization of
macula and optic disc resilient to presence of
disease signs
Detection of Hard exudates, cotton-wool
spots, and identification of circinate clusters
14. 14
Evaluated on 587 images: Sensitivity of 90%
International Clinical Diabetic Macular
Edema Disease Severity Scale
15. 15
Redesign: Module for
Proliferative DR
Diabetes
Proliferative
DR (PDR)
Non-proliferative
DR (NPDR)
• Neovascularization
• Vitreous hemorrhage
• Retinal detachment
Pre-proliferative DR
/ Severe NPDR
Vision loss
• Microaneurysms
• Small hemorrhages
• Exudates
• Macular edema
• Ischemic regions, …
16. 16
Module for proliferative
Diabetic Retinopathy
Module developed with heat map to identify
new vessels /Proliferative vascular abnormality
and retinal detachment
17. 17
Module developed for PDR
identification
Image-level decision of PDR presence
Divide image into
non-overlapping
uniform size
patches
Characterize
local texture
Derive vascular
morphometric
features
Compute a
sensitive
vessel map
For
each
patch
Dense
descriptor of
patch
Recognize
neovascularity
1
2
3
4
5
19. 19
Summary
Since diabetic macular edema accounts for more
than 40% of all DR related vision loss, special
module was developed and evaluated
Misdiagnosis of late stage DR is highly
unfavorable and associated vision loss, so
detection of PDR was developed, identifying
NVE, NVD, Fibro vascular proliferation and
retinal detachment
20. 20
Intelligent identification of image gradability
is necessary for the other modules to be
effective, so gradability module was
developed
Algorithm - good sensitivity and specificity
to detect presence or absence of DR
Cost effective large scale screening of
diabetic patients to prevent blindness in the
population
Summary