Machine Learning based Retinal Therapeutic for Glaucoma

Send Orders for Reprints to reprints@benthamscience.ae
Current Medical Imaging Reviews, 2017, 13(1), pp. 221-226 221
221-227/13 $58.00+.00 © 2017 Bentham Science Publishers
Machine Learning based Retinal Therapeutic for Glaucoma
J.Rubya
, P.S.Jagadeesh Kumarb
, J.Lepikac
, J.Tisad
and J.Nedumaane
a
Research Scholar, Department of Medical Sciences, University of Oxford, United Kingdom
b
Biomedical Engineering Research Centre, Nanyang Technological University, Singapore
c, d, e
Malco Vidyalaya Matriculation Higher Secondary School, Mettur Dam, Tamil Nadu, India
Abstract: Glaucoma is a corpus of visual malady exemplified by the scheduled optic nerve
neuropathy; means to the increasing dwindling in vision ground, resulting in loss of sight.
In this paper, an innovative support vector machine based retinal therapeutic for
glaucoma using adaptive learning algorithm is conservative.
The algorithm has suitable realism; after that sustained-on correlation clustering mode, it
visualizes perfect computations in the multi-dimensional space. Support vector clustering
turns out to be akin to the scale-space advance that investigates the cluster group by
means of a kernel density estimation of the likelihood distribution, where cluster
midpoints are idiosyncratic by the neighborhood maxima of the concreteness.
The prophesied procedure has 94% attainment rate on data set deterrent on a
consolidation of 700 realistic images of resolute and glaucoma retina; therefore, the
computational benefit of depending on the cluster overlapping system pedestal on
adaptive learning algorithm have complete performance in glaucoma therapeutic.
Keywords: Machine learning algorithm, correlation clustering mode, cluster overlapping technique, glaucoma, kernel density estimation,
retinal therapeutic.
1. INTRODUCTION
Support vector machines (SVM) are supervised learning
facsimile with related learning algorithms that examine data
employed for classification and deterioration study. Provided
a set of training instances, each discernible for fitting into
one of two groups, an SVM training algorithm hypotheses a
representation that consigns novel exemplars into one group
or the supplementary, building a binary linear classifier. An
SVM is a depiction of the paradigm as summits in space,
plotted so that the demonstration of the disconnect groups
are alienated by an obvious fissure that is as spacious as
probable. Novel cases are then chronicled into that identical
space and envisaged to fit into a group pedestal on which
plane of the fissure they descend on. In adding up to
accomplishing linear taxonomy, SVM can shrewdly realize a
non-linear taxonomy making use of what is portrayed as
kernel trick, completely recording their inputs into multi-
dimensional typical spaces. When statistics are not
recognized, supervised learning is not feasible, and an
unsupervised learning practice is unpreventable, which
endeavours to locate characteristic clustering of the
information to groups, and then plot novel information to
these fashioned groups. The clustering algorithm which
affords an enhancement to the SVM is called support vector
clustering and is often employed in engineering relevance
when information is not recognized or when only certain
information is recognized as a pre-processing for a taxonomy
pass.
2. GLAUCOMA
Glaucoma is an anthology of visual disorder exemplified
by the scheduled optic nerve neuropathy; manner to the
increasing dwindling in vision ground, resulting in loss of
sight. The dialect "glaucoma" is from Greek glaukos which
refers sapphire, emerald, or hoary [1] [2]. Glaucoma is a
collection of eye ailment which effect in harm to the optic
nerve and loss of vision [18]. The widespread kind is open-
angle glaucoma through fewer common kinds as well as
closed-angle and normal-tension glaucoma. Open-angle
glaucoma extends gradually with time and there is no twinge

222 Current Medical Imaging Reviews, 2017, Vol. 13, No. 1 J.Ruby et al.
© 2017 Bentham Science Publishers
[10] [19]. Side vision may embark on to diminish pursued by
innermost vision ensuing in loss of vision if not diagnosed
precisely [6] [26]. Closed-angle glaucoma can present slowly
or abruptly [14]. The rapid staging might engage rigorous
eye twinge, imprecise vision, mid-dilated pupil, reddishness
of the nausea. Loss of vision due to glaucoma [19], once has
emerged, is perpetual. Risk factor for glaucoma embraces
increased pressure in the eye as illustrated in Figure.1, a
family evidence of the fact, migraines, blood pressure, and
corpulence. In support of eye pressure, a significance of
larger than 21 mmHg is frequently employed [18], with
constrained pressure leading to a superior risk [3]. However,
a small number might comprise high eye pressure for long
time and by no means makes bigger harm. On the other
hand, optic nerve injury might arise with normal pressure
too, accredited as normal-tension glaucoma [12].
Figure. 1 Eye with Glaucoma
The open-angle glaucoma is supposed to be sluggish egress
of aqueous humor from side to side the trabecular meshwork
whereas in closed-angle glaucoma the iris obstructs the
trabecular meshwork. Exploration is by a long-drawn-out
eye assessment. Often the optic nerve reveals an attribute
recognized as cupping. If treated near the beginning it is
likely to premeditated or put a stop to the evolution of
ailment with medicine, laser healing, or surgical procedure.
The objective of these healings is to diminish eye pressure
[14]. A large variety of dissimilar modules of glaucoma
medication are accessible. Laser healing may be effectual in
both open-angle and closed-angle glaucoma. Huge categories
of glaucoma surgeries might be employed in people who do
not retort adequately to further procedures. Ruling of closed-
angle glaucoma is a therapeutic crisis. Concerning 69 to 122
million populaces have glaucoma globally. The disease
influences regarding 30 million citizens in India. It transpires
more frequently among elder citizens. Closed-angle
glaucoma is further rampant in women. Glaucoma is
baptized the "quiet burglar of vision" since the loss of vision
typically happens above an extended epoch of time.
Globally, glaucoma is the second-leading reason for the loss
of vision following cataracts [4]. Examining for glaucoma is
typically carried out as fraction of a customary eye
transmission accomplished by optometrists and
ophthalmologists. Checking for glaucoma ought to comprise
dimensions of the tonometry, gonioscopy, and assessment of
the optic nerve to seem for any perceptible harm to it, or
modify in the cup-to-disc ratio and as well rim facade and
vascular alteration. A ceremonial visual field test is supposed
to be attained [15]. The retinal nerve fiber covering can be
appraised with imaging methods such as optical tomography,
polarimetry, and ophthalmoscopy. Due to the compassion of
all techniques of tonometry to corneal width, schemes such
as Goldmann tonometry must be amplified with pachymetry
to evaluate Central Corneal Thickness (CCT). A thicker-
than-average cornea can affect in a pressure interpretation
upper than the factual pressure, while a thinner-than-average
cornea can create a pressure reading inferior than the factual
pressure [22] [23]. Since pressure computing error can be
initiated by additional than CCT i.e., corneal hydration,
elastic properties, etc., it is impractical to adjust pressure
computation buoyed only on CCT assessments [18] [20].
The frequency doubling illusion can also be exercised to
categorize glaucoma with the exploit of a frequency
doubling expertise [5].
Inspection for glaucoma also may conceivably be appraised
with further attention specified to gender, pursuit, and
narration of drug employed, refraction, bequest and
ancestor’s healthiness record [24] [27]. Absolute glaucoma is
the final arena of all category of glaucoma [28]. The eye has
rebuff to visualization, lack of pupillary light reflex and
pupillary response, and has a pebbly manifestation. Brutal
throbbing is subsisting in the eye [29]. The therapeutic of
absolute glaucoma is a preparation like cyclophoto-
coagulation, before vaccination of 99% alcohol. Glaucoma is
a parasol expression for eye circumstances which ruin the
optic nerve, and be capable of escorting to blindness. The
foremost reason of impairment to the optic nerve is
Intraocular Pressure (IOP), disproportionate fluid pressure
inside the eye, which might be owing to a variety of rationale
together with obstruction of drainage canals, and tapering or
finality of the slant amid the iris and cornea. The principal
dissection in classifying various kinds of glaucoma is open-
angle and closed angle glaucoma [6]. The open angle relates
to the angle where the iris congregates the cornea which
likely to be as broad and unbolt as it is supposed to be,
permitting the fluid from within the eye to deplete, thus
mitigating the interior pressure. When this angle is
diminished, or congested, pressure can surge, and eventually
harm the optic nerve instigating loss of vision. Chronic
glaucoma relates to slow bottleneck of the drainage canals
succeeding in augmented eye pressure which reasons optic
nerve break [7]. This apparent as a plodding loss of the
visual field, preliminary with a loss of peripheral vision, but
finally the complete vision will be mislaid if not diagnosed.
This is the frequent kind of glaucoma, related for 70% of
cases in India. Inception is sluggish and simple, and loss of
vision is steady and irreparable [25].
Narrow angle glaucoma the iris distorts onward, tapering the
angle that saps the eye, rising pressure inside the eye. If not
diagnosed, it can guide to the medical disaster of angle
closure glaucoma. In angle closure glaucoma, the iris distorts
onward and grounds corporeal contact amid the iris and
trabecular meshwork, which in order wedges depletion of the
aqueous humor from inside the eye. This contact might
intermittently impair the strenuous function of the meshwork
in anticipation of its short fall to maintain rapidity with
aqueous construction, and the intraocular pressure increases
[7]. Beginning of warning sign is impulsive, and roots
throbbing and further indications that are obvious, and is

Machine Learning based Retinal Therapeutic for Glaucoma Current Medical Imaging Reviews, 2017, Vol. 13, No. 1 223
delighted as medical crisis. Nothing like open-angle
glaucoma, angle-closure glaucoma is an outcome of the
angle among the iris and cornea closing. These have a
propensity to happen in the far-sighted, which have slighter
than usual frontal cavity, building the corporal contact
further probable. Normal Tension Glaucoma (NTG) is a
situation where the optic nerve is injured though Intraocular
Pressure (IOP) is in normal choice (12-22mm Hg). At
privileged jeopardy are those with ancestor’s record of NTG,
those of Indian origin, and those with account of heart
disease. The aim of NTG is anonymous. Secondary
glaucoma relates to some crate in which an added syndrome,
distress, medicine or practice grounds augmented eye
pressure, following in optic nerve harm and vision loss, and
might be placid or brutal. It can be owing to eye damage,
irritation, protuberance, or sophisticated cases of diabetes
[8]. It can also be grounded by convinced drugs like steroids.
Healing is likely to be open-angle then angle-closure
glaucoma. In pseudo exfoliation glaucoma (PEX) the
pressure is appropriate to the accretion of infinitesimal
grainy protein filaments, which can chunk usual drainage of
the aqueous humor. PEX is customary in those exceeding 70,
and further in women. Pigmentary glaucoma is rooted by
pigment cells bogging off from the flipside of the iris and
hovering roughly in the aqueous humor. Eventually, these
pigment cells know how to hoard in the frontal cavity in
such a means that it can commence to obstruct the trabecular
meshwork. An unusual circumstance, it transpires typically
between Caucasians, frequently males in their middle 19s to
37s, mainly shortsighted. Primary juvenile glaucoma is a
neonate or infantile aberration where optical hypertension is
perceptible at detention or soon after that and is originated
by anomalies in the frontal cavity angle growth that wedge
depletion of the aqueous humor [17]. Uveitic Glaucoma is
owed to uveitis, the engorgement and irritation of the uvea,
the core stratum of the eye. The uvea offers much of the
blood contribution to the retina. Amplified eye pressure in
uveitis can affect from the soreness by itself or commencing
from the steroids exercised to cure it [9].
3. CLUSTERING ALGORITHM
A support vector depiction of a data set is applied as the
source of the Support Vector Clustering (SVC) algorithm
[11] [13] [16]. Consider { } ⊆ be a data set of N
summits, by means of ⊆I d, the data space [26]. By
means of a nonlinear transformation Φ from to various
high dimensional characteristic spaces, the nominal
surrounding sphere of radius is looked for. This is
portrayed by the constrictions:
‖ ( ) ‖ (1)
where || · || is the Euclidean standard and is the middle of
the sphere. Soft constrictions are integrated by totaling
relaxed variables :
‖ ( ) ‖ (2)
through ≥ 0. To explain this predicament, Lagrangian is
∑ ‖ ( ) ‖
∑ + C∑ (3)
where ≥ 0 and ≥ 0 be Lagrange multipliers, C is an
invariable, and C is a significant term. Placing to zero the
plagiaristic of L with deference to , and , in that order,
guides to
∑ (4)
∑ ( ) (5)
(6)
The complementarily situations of Fletcher (1987) effects in
(7)
‖ ( ) ‖ (8)
It trails from equation (3) that the image of a point
with > 0 and > 0 deceits external to the feature-
space sphere. Equation (4) affirms that such a position has
= 0, hence it can be concluded from Equation (6) that =
C. This will be termed as a Bounded Support Vector or BSV.
A point with = 0 is plotted within or to the facade of the
characteristic space sphere. If its 0 < < C in that context,
equation (8) necessitates that its depiction ( ) pretenses
on the peripheral of the characteristic space sphere. Such a
position will be related to as a support vector or SV. SVs
recline on cluster restrictions, BSVs lounge exterior the
precincts, and all further positions stretch out within them.
Note that however C ≥ 1 no BSVs subsist because of the
constrictions. By means of these relations the variables ,
and may be eliminated, turning the Lagrangian into
the Wolfe dual structure that is a utility of the variables :
∑ ( ) ∑ ( ) (9)
because the variables do not materialize in the Lagrangian
they may be reinstated with the constrictions:
(10)
The SV technique is pursued and symbolizes the dot
products ( ) by a pertinent Mercer kernel
( ). All the way through this manuscript, Gaussian
kernel is employed.
( ) ‖ ‖
(11)
Through width parameter q, as noted in Tax and Duin
(1999), polynomial kernels do not yield taut form
representations of a cluster. The Lagrangian is at present
given by:

∑ ( ) ∑ ( ) (12)
At every position is defined as the aloofness of its
depiction in typical space from the middle of the sphere:
‖ ‖ (13)
In observation of Equation (12) and the implication of the
kernel is rewritten as:
∑ ( ) ∑ ( )
(14)
The radius of the sphere is:
{ | } (15)
The outlines that enfold the positions in data space are
definite by the set
{ | } (16)
They are construed as forming cluster restrictions. In sight of
Equation (14), SVs recline on cluster restrictions, BSVs are
exterior, and every further summit lounge within the clusters.
4. CLUSTER OVERLAPPING ALGORITHM
Cluster overlapping algorithm might be functional in
cases where clusters strappingly overlap; on the other hand, a
diverse elucidation of the corollary is compulsory. This
manuscript is anticipated to exercise in such a case, a high
BSV system, and understand the sphere in characteristic
space as in lieu of cluster interiors, moderate than the shroud
of all data. Note that equation (14) for the manifestation of
the sphere in data space can be articulated as;
{ | ∑ ( ) } (17)
Where ρ is strong-minded by the worth of this summation on
the support vectors. The set of summits together with this
form is given by:
{ | ∑ ( ) } (18)
In the excessive case when almost all data positions are
BSVs (p → 1), the computation in this term,
∑ ( ) (19)
is generally equivalent to
∑ ( ) (20)
This preceding term is familiar as Parzen window estimation
of the density gathering up to an optimization aspect, if the
kernel is not suitably optimized. In this high BSV system,
the form in data space is probable to include a little quantity
of summits which recline close to the paramount of the
Parzen estimated density. In further vocabulary, the form
identifies the middle of the likelihood distribution. SVC is
exploited as a “discordant” clustering algorithm, preliminary
from a minute assessment of q and escalating it. The
preliminary assessment of q might be preferred as
‖ ‖
(21)
At this level, all sets of positions fabricate a considerable
kernel assessment, ensuing in a cluster. At this rate no
outliers are desired, therefore C = 1 is chosen. Since q is
augmented, it is anticipated to discover divergence of
clusters. Albeit seeming as hierarchical huddling, counter
examples are initiated when exercising BSVs. Therefore,
stern hierarchy is not assured, except the algorithm is
functional discretely to every cluster that is moderate than
the whole dataset. This option is not accomplished now, in
turn to demonstrate how the cluster composition is tattered
as q is augmented. Initiating with p = 1/N, or C = 1, any
outliers is not permitted. Proviso, since q is being
augmented, clusters of particular or certain positions break
off, otherwise cluster restrictions befall to be uneven, p
ought to be augmented in turn to scrutinize what ensues
when BSVs are permitted. In common, a superior standard
appears to be based on the quantity of SVs: a least number
ensures smooth precincts. As q enhances this quantity
augments, the same as in Figure. 2. If the extent of SVs is
intense, p ought to be augmented, whereby numerous SVs
might be bowed into BSVs, and smooth cluster restrictions
materialize. On the other hand, it is projected to analytically
enhance q and p beside the direction that assures a negligible
quantity of SVs. A subsequent standard for superior cluster
resolution is the constancy of cluster obligation over some
assortment of the two strictures. An imperative concern in
the discordant loom is the pronouncement when to stop
isolating the clusters. A lot of advances to this predicament
subsist, like Milligan and Cooper (1985), BenHur (2002).
Nevertheless, it is alleged that in SV background it is usual
to exercise the quantity of support vectors as a suggestion of
a consequential resolution. Therefore, it is believed to
impede SVC when the portion of SVs surpasses certain
threshold.
Figure. 2 Overlap between clusters underlying likelihood
distribution
5. MACHINE LEARNING ALGORITHM
The structural SVM algorithm offers a broad scaffold for
erudition with versatile controlled yield spaces [21]. The
learning algorithm exercised in this manuscript is SVM
supervised clustering. The SVM algorithm resolves this
quadratic series:

Machine Learning based Retinal Therapeutic for Glaucoma Current Medical Imaging Reviews, 2017, Vol. 13, No. 1 225
‖ ‖ ∑ (22)
( ) ( ) ( ) (23)
At this point, Equation (22) restrains the distinguishing SVM
quadratic principle and flaccid limits. Disparity in Equation (23)
articulates the pair of restraints that permits to discover the
preferred supposition. This meticulous QP is described the
SVM ∆1m program, that is, flaccid norm is 1, and trouncing
proceeds as the periphery. Added QPs are depicted but SVM
∆1m is utilized because it is further compatible with the
correlation clustering algorithm ( ) signifies a factual
esteemed trouncing among a factual cluster and an expected
cluster . ( ) if = , and ∆ ascends as the two
clusters befall further disparate. In examination segment, two
trouncing functions ∆ are betrothed: a loss pedestal on the
precision and recall gain and a “setwise” loss that reckons the
magnitude of setwise cluster association with the clusters
diverge on. The ( ) function precedes a mutual attribute
depiction of an effort and production . In the erudition for
correlation clustering,
( ) | | ∑ ∑ (24)
because ( ) is the correlation clustering intent, for
each instruction example ( ), and each probable mistaken
clustering , SVM clustering algorithm stumble on the
vector to construct the importance of the intention for the
accurate grouping be superior than the significance of the goal
for this erroneous grouping by no less than a fringe of
supervised clustering with SVM the loss between and . The
advance in the SVM algorithm is to establish with no
constrictions, and iteratively discover the major desecrated
constriction.
6. IMPLEMENTATION AND ANALYSIS
The support vector clustering implemented on the retinal
data set which is a distinctive yardstick in pattern recognition
prose. The iris dataset includes 300 factual images of gritty
and glaucoma retina. Individual cluster is linearly discrete
from the further by an apparent contravene in the likelihood
distribution. The residual clusters have considerable overlap,
and were estranged at q=2 p=0.43. However, at these
standards of the constraints, the auxiliary cluster divides into
two. When these two clusters are well thought-out jointly, it
effected in one miss-classification. Accumulating the third
principal component, accomplished the three clusters at q =
5p = 0.60, with three miss-classifications. Through the fourth
principal component the significance of miss-classifications
augmented to six, employing q=9 p=0.34. Additionally, the
quantity of support vectors augmented with ever-increasing
dimensionality. The enhanced recital in 2D or 3D might be
recognized to the noise attenuation of Principal Component
Analysis. The domino effect is comparably useful with
existing clustering algorithms. Intended for high dimensional
datasets, the predicament was to get hold of a support vector
description: the quantity of support vectors shoot from one
cluster to all data points in a separate cluster.
The quadratic programming quandary of Equation (18) can
be resolved by the cluster overlapping algorithm which was
projected as an adept contrivance for clusters that strappingly
overlap. A heuristic is employed to subordinate this estimate:
the complete adjacency matrix is not premeditated, but only
the adjoining with support vectors. The memory necessities
of the cluster overlapping algorithm are squat: it can be
realized by means of O (1) memory at the expenditure of a
decline in competence. This formulates SVC practical even
for awfully outsized datasets. Though while sprint on the
NP-coreference predicament, following that learning
algorithm congregate in relation to 1121precincts, which be
conventional into an SVM QP reoptimized 50 times. The
transparency of clustering with these diminutive sets is small
comparative to the time expend in reoptimizing the QP; by
means of Gaussian kernel, merely one percent of the time
depleted in reoptimizing the QPs was exhausted for
clustering. Of all the accounted testing, the preeminent SVM
learning algorithm continually took less time to congregate
with below a more distinctive instance.
CONCLUSION
Machine learning based glaucoma curative employing
support vector clustering algorithm is anticipated. This course of
action has no unequivocal presumption of both the quantity, or
the silhouette of clusters. It has two constraints, allowing it to
attain assorted clustering responds. The constraint q of the
Gaussian kernel establishes the extent at which the information
is explored, and as it is augmented, clusters instigate to
segregate the customary and retina with glaucoma. The other
constraint, p, is the malleable periphery invariable that pedals
the quantity of outliers. This constraint facilitates scrutinizing
glaucoma exaggerated eye and untying among overlapping
clusters. This is in distinction with a large amount of clustering
algorithms ascertained in the prose that have rebuff coordination
for dealing with outliers. Nevertheless, it might be outstanding
for clustering glaucoma instances with strappingly overlapping
clusters; SVC might demarcate only comparatively diminutive
cluster nucleus. A surrogate for overlapping clusters is to make
use of a support vector description for all clusters. An
exceptional benefit of this algorithm is that it can engender
cluster restrictions of subjective contour, while existing
algorithms that utilize the geometric expression are the largest
part habitually restricted to hyper-ellipsoids. In this deference,
SVC is redolent of highly categorized neurons discrete in a
multi-dimensional attributes pace. The anticipated algorithm has
a deviating improvement; being pedestal on kernel technique it
evades unambiguous computations in the multi-dimensional
attribute space, and so is further competent. In the high p system
SVC turns out to be comparable to the scale-space advance that
investigates the cluster organization by means of a Gaussian
Parzen window approximation of the likelihood distribution,
where cluster midpoints are distinctive by the neighborhood
maxima of the concreteness. The anticipated scheme has 90%
attainment rate on data set restraining a concoction of 300
factual images of gritty and glaucoma retina and therefore the
computational benefit of depending on the SVM quadratic
normalization have introverted comprehensive resolution in
performing glaucoma dexter.

REFERENCES
[1] Qi Wang, Sinisa D. Grozdanic, Matthew M. Harper, Nicholas
Hamouche, Helga Kecova, Tatjana Lazic, Chenxu Yu. Exploring Raman
spectroscopy for the evaluation of glaucomatous retinal changes. Journal of
Biomedical Optics, 16(10), 107006; pp.1-9, October 2011.
[2] Apeksha R. Padaria, Bhai lal Limbasiya. A Review Paper on Detection
of Optic Disc Damage using Retinal Images. International Journal of
Computer Applications, Vol.111, No.13; pp.1-4, 2015.
[3] Syed Akhter Hussain, Holambe A. Automated Detection and
Classification of Glaucoma from Eye Fundus Images: A Survey.
International Journal of Computer Science and Information Technologies,
Vol.6, No.1; pp.1217-1224, 2015.
[4] Imran Qureshi. Glaucoma Detection in Retinal Images Using Image
Processing Techniques: A Survey. International Journal of Advanced
Networking and Applications, Vol.7, No.2; pp.2705-2718, 2015.
[5] Darsana S, Rahul M Nair. A Novel Approach towards Automatic
Glaucoma Assessment. International Journal of Scientific Research
Engineering and Technology, Vol.3, Issue.2; pp.281-288, 2014.
[6] P.S.Jagadeesh Kumar, J.Ruby, J.Lepika, J.Tisa, J.Nedumaan. Glaucoma
Detection and Image Processing Approaches: A Review. Journal of Current
Glaucoma Practice, Vol. 8, Issue 1, January 2014, pp.36-41, 2014.
[7] Archana Nandibewoor, S B Kulkarni, Sridevi Byahatti, Ravindra
Hegadi. Computer Based Diagnosis of Glaucoma using Digital Fundus
Images. Proceedings of the World Congress on Engineering, Vol III,
WCE’2013, July 3-5, London, United Kingdom, 2013.
[8] Malaya Kumar Nath, Samarendra Dandapat. Techniques of Glaucoma
Detection from Color Fundus Images: A Review. I.J. Image, Graphics and
Signal Processing, Vol.4, No.9; pp.44-51, 2012.
[9] P.S.Jagadeesh Kumar et al.Deordination of Retinal Ingerence Repugnant
on Support Vector Machine for Glaucoma Dexter. International Journal of
Biomedical and Biological Engg., World Academy of Science, Engineering
and Technology, Vol.3, No.7, 2015.
[10] K.Narasimhan, K.Vijayarekha. An efficient automated system for
glaucoma detection using fundus image. Journal of Theoretical and Applied
Information Technology, Vol.33, No.1; pp.104-110, 2011.
[11] S. Wang, Z. Li, W. Chao, and Q. Cao. Applying adaptive oversampling
technique based on data density and cost-sensitive SVM to imbalanced
learning. Proc. of The 2012 International Joint Conference on Neural
Networks (IJCNN), 10-15, June 2012, Brisbane, Australia.
[12] N.B.Prakash, D.Selvathi. An efficient detection system of screening
glaucoma in retinal images. Biomedical & Pharmacology Journal, Vol.10,
No.1; pp.459-465, 2015.
[13] S. Lahmiri, C. S. Gargour, and M. Gabrea. An EMD-SVM screening
system for retina digital images: the effect of kernels and parameters. Proc.
of 11th International Conference on Information Sciences, Signal
Processing and their Applications (ISSPA), 2-5, July 2012, Canada.
[14] Apeksha Avinash, K. Magesh, C. Vinoth Kumar. Sift feature based
detection of glaucoma. Proceedings of IRF International Conference, 29-30,
October 2015, Chennai, India.
[15] Zhang Z, Khow CK, Liu J, Cheung YLC, Aung T, et al. Automatic
Glaucoma Diagnosis with mRMR-based Feature Selection. Journal of
Biomet Biostat S7:008; pp.1-8, 2012.
[16] Abhishek Dey, Samir K. Bandyopadhyay. Automated Glaucoma
Detection Using Support Vector Machine Classification Method. British
Journal of Medicine and Medical Research, 11(12); pp.1-12, 2015.
[17] Ganesh Babu T. R, R. Sathish Kumar, Rengaraj Venkatesh.
Segmentation of Optic Nerve Head for Glaucoma Detection using Fundus
images. Biomedical & Pharmacology Journal, Vol. 7(2); pp.697-705, 2014.
[18] Gauri Borkhade, Ranjana Raut. Support Vector Machine Neural
Network Based Optimal Binary Classifier Diabetic Retinopathy.
International Journal of Innovative Research in Computer and
Communication Engineering, Vol.3, Issue.1; pp.136-141, 2015.
[19] Srinivasan C, Suneel Dubey, Ganesh Babu T.R. Complex texture
features for glaucoma diagnosis using support vector machine. International
Journal of MC Square Scientific Research, Vol.7, No.1; pp.84-90, 2015.
[20] Dimitrios Bizios, Anders Heij, Jesper Leth Hougaard, Boel Bengtsson.
Machine learning classifiers for glaucoma diagnosis based on classification
of retinal nerve fibre thickness parameters measured by Stratus OCT.
Journal of Acta Ophthalmol, Vol. 88; pp.44–52, 2010.
[21] K. Stapor. Support vector clustering algorithm for identification of
glaucoma ophthalmology. Bulletin of the Polish Academy of Sciences,
Technical Sciences, Vol. 54, No.1; pp.139-141, 2006.
[22] J. Liu, D.W.K. Wong, J.H. Lim, X. Jia, F. Yin, H. Li, W. Xiong, and
T.Y. Wong. Optic cup and disk extraction from retinal fundus images for
determination of cup-to-disc ratio. Proc. of 3rd
IEEE Conference on
Industrial Electronics and Applications (ICIEA), 3-5, June 2008, Singapore.
[23] Syed SR. Abidi, Paul H, Sanjan Yun, Jin Yu. Automated Interpretation
of Optic Nerve Images: Data Mining Framework for Glaucoma Diagnostic
Support. Journal of Med Info, IOS Press; pp.1309-1313, 2007.
[24] Rakhi Ramachandran Nair, Chippy Jacob, Tessy Vincent, Remyamol
R. Survey on Automatic Detection of Glaucoma based on Eye Images.
International Journal of Innovative Research in Science, Engineering and
Technology, Volume 5, Special Issue 14; pp.22-26, 2015.
[25] G. Yang, Y. Ren, Q. Pan, G. Ning, S. Gong, G. Cai, Z. Zhang, and L.
Li, J. Yan. A heart failure diagnosis model based on support vector machine.
Proc. of 3rd
International Conference on Biomedical Engineering and
Informatics (BMEI), 16-18, October 2010, Yantai, China.
[26] P.S.Jagadeesh Kumar et al. Putrefied Retinal Inquisition on Support
Vector Machine for Glaucoma Curative. International Conference of
Computer Science and Engineering (ICCSE), IAENG World Congress on
Engineering, 2015, London, United Kingdom.
[27] S. Zou, Y. Huang, Y. Wang, J. Wang, and C. Zhou. SVM learning from
imbalanced data by GA sampling for protein domain prediction. Proc. of 9th
International Conference for Young Computer Scientists (ICYCS), 18-21,
November 2008, Hunan, China.
[28] Trevor Hastie, Saharon Rosset, Robert, Ji Zhu. The Entire
Regularization Path for the Support Vector Machine. Journal of Machine
Learning Research, 5 (2004); pp.1391–1415, 2004.
[29] P.S.Jagadeesh Kumar, J.Ruby, J.Lepika, J.Nedumaan, J.Tisa.
Healthcare of Diabetes in Monitoring Patient Blood Glucose at Remote
Areas Using Cloud Technology. IOSR Journal of Dental and Medical
Sciences, Vol. 14, Issue 4, pp.108-116, 2015.
Received: March 20, 2015 Revised: April 16, 2016 Accepted: January 12, 2017

Machine Learning based Retinal Therapeutic for Glaucoma

More Related Content

What's hot

Similar to Machine Learning based Retinal Therapeutic for Glaucoma

More from DR.P.S.JAGADEESH KUMAR

Recently uploaded

Machine Learning based Retinal Therapeutic for Glaucoma