Transcript Liu-Poster

Architecture for Automatic
GLaucoma Diagnosis and
Its Genetic Association
Study through Medical
Image InformAtics
(AGLAIA)
Jimmy Liu Jiang1, Zhang Zhuo1, Wong Wing Kee1, Tan
Ngan Meng1, Yin Feng Shou1, Lee Benghai1, Cheng Jun1,
Wong Tien Yin2
1: Institute for Infocomm Research, Agency for Science,
Technology and Research (A*STAR), Singapore
2: Singapore Eye Research Institute
1
AGLAIA Single statement
A system for the automated detection of
glaucoma from multiple image features of
retinal images with possible genetic
analysis.
Glaucoma Has No Cure!
 Glaucoma
 Irreversible loss of optic nerves
 Leading to blindness
 5.7 million glaucoma blind (total 38 million blind in the
world)
 BUT: glaucoma can be slowed if detected early
 Vision loss (functional)
 Optic nerve damage (imaging)
Optic Nerve Damage
Precedes Vision Loss
No Glaucoma Early Late Stage
Optic Nerve
Nerve Damaged
Pictures from NIH
Severe
Vision Loss
Measure Optic Nerve Damage from
Retinal Image - CDR
 CDR (Cup to Disc Ratio) value is an
important indicator for Glaucoma
 Automatic CDR measurement
 Saving clinician’s time
 Results reproducible
 Ideal for mass screening
Cup
Disc
AGLAIA

Step 1: Disc segmentation

Step 2: Boundary smoothing
of the detected disc

Step 3: Cup segmentation

Step 4: Boundary smoothing
of the detected cup

Step 5: Intelligent Fusion
AGLAIA Framework
CDR
Fundus
image
Asymmetry
Notching
Gradeability
Rim
PPA
Vessel
Pattern
Kink
ISNT Rule
Tilted Disc
Hemorrhage
RNFL
Disc Size
Glaucoma risk
analysis Report
Clinical data
Intelligent
fusion
MINING
Genome
data
GWAS
13 Image cues used in AGLAIA
AGLAIA aims to provide a multi-modality system that captures globally a range of
parameter that is indicative of early glaucoma damage. AGLAIA will automatically
measure and assess the following features objectively and quantitatively, features
which are currently evaluated subjectively by glaucoma specialist in clinical practice:













Cup-to-disc ratio (CDR) – further refinement will be made based on initial algorithms
developed in ARGALI
Disc hemorrhage (DH)
Thinning of the NRR (NeuroRetinal Rim)
Notching of the NRR
Compliance of NRR width by ‘ISNT Rule’ (inferior≥superior≥nasal≥temporal)
Inter-eye asymmetry
Parapapillary atrophy (PPA)
Blood vessel pattern analysis
Blood vessel kink analysis
Tilted Disc - quantify the degree of tilting based on the disc contour
Disc Size - automatically classify disc size to "large, medium or small" categories based on
automatic disc measurement
Gradeability – analyze the image and determine gradeability
Check the presence of RNFL (Retinal Nerve Fiber Layer) defect.
Image Datasets
 Datasets used for testing
 Retinal Vessel and Glaucoma Subtype Study
(RVGSS)
 65 glaucoma images
 Singapore Indian Chinese Cohort (SICC)
 224 non-glaucoma images
 Clinical CDR provided
Test Result Summary
 Specificity
 0.88
 Sensitivity
 0.95
 Average absolute CDR error (wrt clinical
CDR)
 0.14
AGLAIA Market Potential
 The Technology can be readily implemented in
currently available instruments for ocular
screening without extensive modifications.
 The Technology could potentially be of interest
to ocular instrument makers to incorporate the
Technology into their equipment, and to health
institutions such as clinics and hospitals for
glaucoma screening.
AGLALI CDR Absolute Error Distribution
Error Interval
Receiver Operating Characteristic (ROC)
Curve
AUC
Sensitivity
Specificity
ICC
Maximum
Std.
Deviation
0.94
0.954
0.88
0.553
0.65
0.12