diagnostics

Article
Automatic Detection of Age-Related Macular Degeneration
Based on Deep Learning and Local Outlier Factor Algorithm
Tingting He , Qiaoer Zhou and Yuanwen Zou *

College of Biomedical Engineering, Sichuan University, Chengdu 610065, China;

2019223010062@[Link] (T.H.); zhou_qiaoer@[Link] (Q.Z.)
* Correspondence: zyw@[Link]

Abstract: Age-related macular degeneration (AMD) is a retinal disorder affecting the elderly, and
society’s aging population means that the disease is becoming increasingly prevalent. The vision in
patients with early AMD is usually unaffected or nearly normal but central vision may be weakened
or even lost if timely treatment is not performed. Therefore, early diagnosis is particularly important
to prevent the further exacerbation of AMD. This paper proposed a novel automatic detection method
of AMD from optical coherence tomography (OCT) images based on deep learning and a local outlier
factor (LOF) algorithm. A ResNet-50 model with L2-constrained softmax loss was retrained to extract
features from OCT images and the LOF algorithm was used as the classifier. The proposed method
was trained on the UCSD dataset and tested on both the UCSD dataset and Duke dataset, with an
accuracy of 99.87% and 97.56%, respectively. Even though the model was only trained on the UCSD
dataset, it obtained good detection accuracy when tested on another dataset. Comparison with other
methods also indicates the efficiency of the proposed method in detecting AMD.

Keywords: age-related macular degeneration; optical coherence tomography; deep learning; local



outlier factor


Citation: He, T.; Zhou, Q.; Zou, Y.
Automatic Detection of Age-Related
Macular Degeneration Based on
1. Introduction
Deep Learning and Local Outlier
Factor Algorithm. Diagnostics 2022, Age-related macular degeneration (AMD) is the fourth most prevalent ocular disease
12, 532. [Link] resulting in vision loss in the macula [1]. The macula is located in the optical center of
diagnostics12020532 the human eye and is an important part of the retina. It is required for reading, driving,
watching TV, and performing many other daily activities [2]. Of all cases of blindness
Academic Editor: Jae-Ho Han
worldwide, 8.7% are caused by AMD and the number of patients with AMD was estimated
Received: 19 January 2022 at around 196 million in 2020, predicted to rise to 288 million by 2040 [3].
Accepted: 17 February 2022 AMD is broadly classified into non-exudative or dry AMD and exudative or wet AMD.
Published: 18 February 2022 The difference between dry and wet AMD is that dry AMD does not have any blood or
Publisher’s Note: MDPI stays neutral
serum leakage. Around 85% to 90% of AMD cases are dry [4]. Patients suffering from
with regard to jurisdictional claims in
dry AMD have a significant anomaly known as drusen in the retinal pigment epithelium
published maps and institutional affil- (RPE) layer. The formation of drusen leads to a thinning and drying out of the macula,
iations. which results in the loss of macular function. Although patients with dry AMD may still
have a good central vision, they may have significant functional limitations, including
limited night vision, vision fluctuations, and reading difficulties due to a limited area
of central vision. Moreover, a certain percentage of dry AMD may develop into wet
Copyright: © 2022 by the authors. AMD as time goes by [5]. In wet AMD, patients may see dark spots in their central
Licensee MDPI, Basel, Switzerland. vision due to blood or fluid leakage under the macula. The main pathogenesis of wet
This article is an open access article AMD is choroidal neovascularization (CNV), which occurs under the retina and macula.
distributed under the terms and This neovascularization may lead to macular swelling and a reversible loss of vision, or
conditions of the Creative Commons bleeding, which can be highly toxic to the overlying photoreceptors, sometimes even
Attribution (CC BY) license (https://
causing irreversible vision loss [6,7]. In wet AMD, vision loss may be rapid and progressive.
[Link]/licenses/by/
Once CNV has developed in one eye, the other eye is in a high-risk state and requires
4.0/).

Diagnostics 2022, 12, 532. [Link] [Link]

Diagnostics 2022, 12, 532 2 of 13

periodic eye examination [8]. Therefore, regular screening of the retina is crucial for the
diagnosis and treatment of AMD and the prevention of further deterioration.
Ophthalmologists can detect AMD-related lesions through a variety of methods with
the continuous development of imaging technology [9–12]. Optical coherence tomog-
raphy (OCT) uses the basic principle of a weak coherence interferometer to carry out
high-resolution cross-sectional tomography of the internal microstructure of materials or bi-
ological systems [13]. It has been widely used in many fields and has great scientific research
and application potential in biomedical, agricultural and industrial detection [14–18]. In
recent years, OCT has become a major tool to diagnose AMD and monitor its progress [19].
It reflects the physical structure of the retina accurately and effectively in a non-contact and
non-invasive way [20,21]. It can clearly describe the particular pathology relating to AMD,
such as drusen, intra-retinal fluid (IRF), sub-retinal fluid (SRF), sub-retinal hyper-reflective
material and RPE detachment [22].
The OCT images to be analyzed increase dramatically with a more widespread screen-
ing of the retina. The number of professional ophthalmologists is limited because training
new professional doctors is a long process. It is often difficult for patients of AMD to
receive timely diagnosis and treatment, especially in places where medical resources are
insufficient. Therefore, a computer-aided diagnosis (CAD) system that can automatically
detect AMD from a large number of retinal OCT images is urgently needed. In recent years,
researchers have launched kinds of studies on AMD detection and classification based on
retinal OCT images with the development and improvement of computer technology and
image processing algorithms [23–34].

2. Related Work
Layer segmentation is crucial in many automatic analysis algorithms based on retinal
OCT images. The position and thickness of each retinal layer are obtained according
to the result of the layer segmentation algorithm, then by analyzing the similarities and
differences between the layer index of the tested image and the reference image, a variety
of issues, including lesion detection and positioning, can be addressed.
Farsiu et al. [23] introduced a semi-automatic segmentation of RPE, RPE drusen
complex (RPEDC) and total retina (TR) boundaries. Then, volumes of TR, RPEDC and
abnormal RPEDC of each subject were measured and compared with the normal thickness
generated by control subjects to detect AMD. The area under the curve (AUC) of the
receiver operating characteristic (ROC) for this classifier was 0.9900.
Naz et al. [24] proposed an algorithm to detect the AMD-effected OCT scans by
calculating the difference between the RPE layer and a second-order polynomial curve.
The method was made time efficient by using an intensity-based threshold method for the
RPE segmentation. A dataset with 25 AMD and 25 healthy images was used, and the study
obtained an accurate detection of AMD with 96.00% accuracy.
Arabi et al. [25] used the binary threshold method to extract the RPE layer, sampled
the extracted layers and counted the number of white pixels in each sample. The mean
value of the numbers of pixels was calculated and classified. They tested the approach on
16 images and obtained an accuracy of 75.00%.
Thomas et al. [26] proposed an algorithm based on RPE layer detection and baseline
estimation using statistical methods and randomization for the detection of AMD from
retinal OCT images. The method was tested on a public dataset including 2130 images and
achieved an overall accuracy of 96.66%.
Sharif et al. [27] presented a method based on feature extraction and the support vector
machine (SVM). First, the RPE layer was extracted by utilizing the graph theory dynamic
programming technique, then a unique feature set consisting of features extracted from the
difference signal of RPE and the inner segment outer segment layer of RPE was obtained.
Finally, the SVM classifier was used to detect AMD-affected images from 950 OCT scans,
and an accuracy of 95.00% was obtained.
Diagnostics 2022, 12, 532 3 of 13

Although the above methods based on layer segmentation obtained promising results,
they are not suitable for large-scale AMD detection. The convolutional neural network
(CNN), which emerged at the end of the 20th century, has significantly improved the ability
to classify images [28].
Lee et al. [29] classified 52,690 normal and 48,312 AMD OCT images utilizing a
modified version of the VGG-16 CNN model and obtained an overall accuracy of 93.40%.
Serener et al. [30] compared two pre-trained CNN, namely AlexNet and ResNet-18, to
automatically classify OCT images for dry and wet AMD diseases, respectively. In both
cases, the ResNet-18 model outperformed the AlexNet model, and the AUC of the ResNet
model for each AMD stage was 0.9400 and 0.9300, respectively.
Thomas et al. [31,32] conducted a number of studies based on AMD detection using
OCT images. In [31], a multiscale and multipath CNN with six convolutional layers
was proposed and finally achieved an overall accuracy of 98.79% with the random forest
(RF) classifier. Later, in [32], they introduced another novel multiscale CNN with seven
convolutional layers to classify AMD and normal OCT images. The multiscale convolution
layer enables a large number of local structures to be generated with various filter sizes.
The proposed CNN network finally achieved an accuracy of 99.73% on the UCSD dataset.
Yoo et al. [33] utilized VGG-19 pre-trained with images from ImageNet as a feature
extractor, and a multiclass RF classifier was operated to detect AMD images. The overall
accuracy using OCT alone was 82.60% on a small dataset including both OCT and matched
fundus images. Kadry et al. [34] extracted handcrafted features, such as the local binary
pattern (LBP), the pyramid histogram of oriented gradients (PHOG), and the discrete
wavelet transform (DWT) from the test images and concatenated them with the deep
features of VGG-16. The proposed technique achieved an accuracy of up to 97.00% for OCT
images with different binary classifiers.
In this study, we presented a novel method for the detection of AMD based on OCT
images and showed it to be more effective than existing methods. The rest of the paper is
structured as follows. The proposed methodology is given in Section 3, then the datasets
used for the experiment and the parameters of the model are given in Section 4. The
experimental results and discussion are shown in Section 5. The conclusion is given in
Section 6.

3. Method
In this study, a two-stage model was proposed for the detection of AMD from OCT
images, as shown in Figure 1. The first stage involved a classification model using a deep
CNN, while in the second stage, an outlier detection method was used for detecting AMD.
In the first stage, a deep CNN based on ResNet-50 [35] was used for classification, and
transfer learning was performed using AMD and normal OCT images [36]. After retraining,
the last layer of the network (classification layer) was removed and the model was regarded
as an image feature extractor.
In the second stage, the normal images in the training set were imported to the network
to obtain a normal image feature vector set. During testing, images in the test set were
imported into the network in turn, and each test image could obtain a corresponding
feature vector. Both the normal image feature vector set and test image feature vector were
used as inputs of the local outlier factor (LOF) algorithm [37]. Finally, the LOF algorithm
classified the test image as normal or abnormal (corresponding to AMD).

3.1. ResNet-50
ResNet was proposed by He et al. [35]. In the deep structure of CNN, as the layers
deepen, gradient disappearance or explosion may occur, resulting in a drop in accuracy.
The problem can be solved by the residual network, improving the performance and
increasing the depth of the network at the same time. ResNet has been widely used
in the field of medical image classification, in applications such as multi-label chest X-
Diagnostics 2022, 12, 532 4 of 13

Diagnostics 2022, 12, x FOR PEER REVIEW 4 of 14

ray classification [38], the diagnosis of COVID-19 [39] and exudate detection in fundus
images [40].

Figure 1. Framework of the proposed approach. Stage 1 is the process of training and Stag
process of feature extraction and classification. The diagram on the far right is a visual re
tion of Stage 2.

3.1. ResNet-50
ResNet was proposed by He et al. [35]. In the deep structure of CNN, as th
deepen, gradient disappearance or explosion may occur, resulting in a drop in a
The problem can be solved by the residual network, improving the performance
creasing the depth of the network at the same time. ResNet has been widely . use
field 1.1.
Figure
Figure ofFramework
medical of
Framework image
of the classification,
the proposed
proposed approach. in applications
approach. Stage
Stage 11 is
is the processsuch
the process of as multi-label
of training
training and
and Stage
Stage 22 is chest
is the
the X-ra
fication
process of [38],
feature the diagnosis
extraction and of COVID-19
classification. The [39]
diagram and
on the exudate
far right is adetection
visual
process of feature extraction and classification. The diagram on the far right is a visual representa- in fundus
representation ima
tion of Stage
of StageIn 2.
2. this work, we utilized a 50-layer structure of ResNet. The residual block s

of ResNet-50
In this work, is
3.1. ResNet-50 weshown
utilized ain50-layer
Figurestructure
2. Theofinput ResNet. x The
is transferred
residual blockacross
structure layers th
ofshortcut
ResNet-50 connection
ResNet was is shown
proposed intoFigure
be added
by He 2.
into
The[35].
et al. inputthex is
In the
output F(x) after
transferred
deep structure across convolution,
layers
of CNN, asthrough
the layers
anda then
put y =connection
shortcut
deepen, F(x) + xdisappearance
gradient is
to obtained.
be added intoorThethe residual F(x)block
outputmay
explosion can fullyintrain
after convolution,
occur, resulting and theinthe
then
a drop underlying
output
accuracy. net
ythe
= F(x) + x
accuracyis obtained. The residual block can fully train the underlying network, so the
The problem cancan be significantly
be solved by the residual improved
network,as the depth
improving the increases.
performance and in-
accuracy can be significantly improved as the depth increases.
creasing the depth of the network at the same time. ResNet has been widely used in the
field of medical image classification, in applications such as multi-label chest X-ray classi-
fication [38], the diagnosis of COVID-19 [39] and exudate detection in fundus images [40].
In this work, we utilized a 50-layer structure of ResNet. The residual block structure
of ResNet-50 is shown in Figure 2. The input x is transferred across layers through a
shortcut connection to be added into the output F(x) after convolution, and then the out-
put y = F(x) + x is obtained. The residual block can fully train the underlying network, so
the accuracy can be significantly improved as the depth increases.

Figure
Figure2. A
2. “Bottleneck” building
A “Bottleneck” block forblock
building ResNet-50,
for ResNet-50, × 1 layers
where the 1 where theare1 responsible
× 1 layersfor
are respo
reducing and then increasing (restoring) dimensions, leaving the 3 × 3 layer with a bottleneck of
reducing and then increasing (restoring) dimensions, leaving the 3 × 3 layer with a bott
smaller input/output dimensions.
smaller input/output dimensions.
Figure 3 shows the architecture of ResNet-50 used in this study. The last fully-
Figure
connected layer3was
shows thetoarchitecture
adjusted binary outputof ResNet-50
classes for AMDused in thisinstead
and normal study. of The
the
Figure 2. A “Bottleneck” building block for ResNet-50, where the 1 × 1 layers are responsible for
last fu
1000 output
nected and
reducing
classes
layer
thenwas
of the ImageNet,
adjusted
increasing
and
to binary
(restoring)
the loss
output
dimensions,
function
classes
leaving
was L2-constrained
the 3 × for AMD
3 layer with aand
softmax
normal
bottleneck of instea
loss [41].
smaller input/output dimensions.

Figure 3 shows the architecture of ResNet-50 used in this study. The last fully-con-
nected layer was adjusted to binary output classes for AMD and normal instead of the
Diagnostics 2022, 12, x FOR PEER REVIEW 5 of 14

Diagnostics 2022, 12, 532 5 ofsoftmax

1000 output classes of the ImageNet, and the loss function was L2-constrained 13

loss [41].

Figure
[Link]
Thearchitecture
architecture ofof the
theResNet-50
ResNet-50 used.
used. L2-constrained
L2-constrained is implemented
is implemented via
via the the L2-Norm
L2-Norm
layer and
layer andthe
theScale
Scalelayer in the
layer in theOUTPUT
OUTPUTmodule.
module.

TheL2-constrained
The L2-constrained softmax
softmaxloss
lossisisgiven byby
given Equation (1) [41]:
Equation (1) [41]:
M WyT f ( xW T by
)+ b yi
1 M y f(x
i )+
minimize − 1M e
e i ii
∑ log i

minimize - 1 log
i=
W T f ( x )+by
∑Cj=1 e yi i T i (1)
xi )||i2=1= α, ∀
C
|| f (M
W f(x )+ b
subject to e 3yi. . . iM yi
i = 1, 2,
j =1
(1)

where xi is the input image subject

in theto || f(xi ) of
mini-batch = α,M,
|| 2size ∀iyi=is1,the2,corresponding
3...M class label,
f(xi ) is the feature descriptor obtained from the penultimate layer, C is the number of classes,
where
and W is the
xi and input
b are image and
the weight in the
biasmini-batch of size
of the last layer. AnM, yi is the L2
additional corresponding class label,
constraint is added
on the basis of the traditional softmax function, and the constraint is enforced
f(xi) is the feature descriptor obtained from the penultimate layer, C is the number of clas- by adding
anand
ses, L2-normalize
W and b are (L2-Norm) layer,and
the weight followed
bias ofbythea scale
last layer,
[Link] Anshown in Figure
additional L2 3. This
constraint is
constraint restricts the feature to a hypersphere with a fixed radius through a parameter,
added on the basis of the traditional softmax function, and the constraint is enforced by
which brings the features from the same class closer to each other and separates the features
adding an L2-normalize (L2-Norm) layer, followed by a scale layer, as shown in Figure 3.
from different classes in the normalized or angular space. In this study, M = 12, C = 2, and
This
α wasconstraint
set to 5. restricts the feature to a hypersphere with a fixed radius through a param-
eter, which brings the features from the same class closer to each other and separates the
3.2. LOF
features Algorithm
from different classes in the normalized or angular space. In this study, M = 12, C
= 2, and α was
The LOF set to 5. was used to divide OCT images into two groups: normal and
algorithm
AMD. The LOF is an outlier detection method that computes the local density deviation
ofLOF
3.2. a given data point with respect to its neighbors [37]. The local density is given by
Algorithm
Equation (2):
The LOF algorithm was used to divide OCT images into two groups: normal and
∑o∈ Nk ( p) reach − distk ( p, o )
lrdk ( pdetection
AMD. The LOF is an outlier ) = 1/( method that computes) the local density deviation
(2)
| Nk ( p)|
of a given data point with respect to its neighbors [37]. The local density is given by Equa-
where lrdk (p) is the local density of object p, Nk (p) is the k-distance neighborhood of p, and
tion (2):
reach-distk (p,o) is the reachability distance of object p with respect to object o, which is given
by Equation (3):
= o1/(
lrd k (p)( p,
 reach − dist (p,o)
o∈N k (p) k
) (2)
reach − dist k ) = max N (p)(o), d( p, o)}
{k − distance
k
(3)

Using lrd (p), the LOF of p is defined by Equation (4):

where lrdk(p) iskthe local density of object p, Nk(p) is the k-distance neighborhood of p, and
reach-distk(p,o) is the reachability distance of object lrd p with(o ) respect to object o, which is given
∑o∈ Nk ( p) lrd k ( p)
by Equation (3): LOFk ( p) = ( k
) (4)
| Nk ( p)|
reach − dist (p,o) = max{k − distance(o),d(p,o)}
where k was set to 20 in this study. kThe output LOF value was used to determine whether p
(3)
was normal or abnormal (corresponding to AMD), as shown in Figure 4.
Using lrdk(p), the LOF of p is defined by Equation (4):
 LOFk (p) = ( )
N k (p)
where k was set to 20 in this study. The output LOF value was used to d
Diagnostics 2022, 12, 532 p was normal or abnormal (corresponding to AMD), as shown in Figur
6 of 13

Figure
Figure 4. 4. Schematic
Schematic illustration
illustration of LOF. LOFof LOF.
can LOF
be used can bethe
to determine used to determine
threshold the thres
of normal OCT
images based on the local density of the normal images in the training set, and discriminate whether
images based on the local density of the normal images in the training set, and d
the test image is normal or abnormal (AMD).
the test image is normal or abnormal (AMD).
4. Experiment
This section consists mainly of two parts: datasets and model training, including
4. Experiment
parameters and the environment.
This section consists mainly of two parts: datasets and model train
4.1. Datasets
rameters and the
The proposed environment.
method was trained and validated using the UCSD dataset [42]. The
UCSD dataset was selected from retrospective cohorts of adult patients between 1 July 2013
and 1 March 2017, and all OCT images were acquired using Spectralis OCT (Heidelberg
4.1. Datasets
Engineering, Heidelberg, Germany) imaging. There were 45,821 AMD images in total,
including
The8616 dry AMD and
proposed 37,205 wet
method wasAMD, and 80%
trained of the
and images wereusing
validated used asthe
the
UCSD
training set and 20% of the images as the validation set, as given in Table 1.
UCSD dataset was selected from retrospective cohorts of adult patie
2013
Table [Link]
Data 1 Marchfor2017,
distribution and
training all OCT
and validation images
using were
the UCSD acquired using Spect
dataset.

bergUCSDEngineering,
Dataset Heidelberg,
AMD Germany) NORMAL imaging. ThereTOTAL were 45,821 A
tal,Training
including
set (80%)8616 dry36,656AMD and 37,205 wet AMD, and
40,912 80% of the im
77,568
Validation set (20%) 9165 10,228 19,393
the training set and 20% of the images as the validation set, as given in
The UCSD test set contains 500 AMD images (250 dry AMD and 250 wet AMD) and
Table 1. Data
250 normal distribution
images. foralso
In addition, we training and
tested the validation
method usingdataset
with the Duke the UCSD dataset.
[43]. The
Duke dataset includes multiple OCT images from 45 subjects (15 dry AMD, 15 DME, and
15 normal) UCSD Dataset
and all images AMD
were acquired in Institutional NORMAL
Review Board-approved protocols
using Spectralis OCT (Heidelberg Engineering Inc., Heidelberg, Germany) imaging. The
proposedTraining set (80%)
method utilized images from AMD and normal 36,656
subjects. 40,912
Validation
The sample imagesset (20%)
in the 91655. Dry AMD is a state10,228
datasets are shown in Figure in which
the macula layer becomes thin and dry. There is a small amount of amorphous material
that aggregates in the cells of the eye, also known as drusen, as shown by the white arrows
The
in Figure UCSD
5. Wet AMD test
refersset contains
to the 500 vessel
irregular blood AMD images
under (250which
the macula, dryisAMD
called and 2
CNV. This blood vessel may cause the macula to rise from its flat position due to fluid
250 normal images. In addition, we also tested the method with the Duk
leakage and bleeding, as shown by the white box in Figure 5a.
Duke dataset includes multiple OCT images from 45 subjects (15 dry A
15 normal) and all images were acquired in Institutional Review Boar
cols using Spectralis OCT (Heidelberg Engineering Inc., Heidelberg, G
The proposed method utilized images from AMD and normal subjects
2, 12, x FOR PEER REVIEW 7 of 14

Diagnostics 2022, 12, 532 7 of 13

CNV. This blood vessel may cause the macula to rise from its flat position due to fluid
leakage and bleeding, as shown by the white box in Figure 5a.

(a)

(b)
Figure 5. Sample images:
Figure (a) belongs
5. Sample to the(a)
images: UCSD dataset;
belongs (b) belongs
to the UCSD dataset;to(b)
the Duketodataset.
belongs The
the Duke dataset. The white
white arrows point arrows
to drusens
pointintothe dry AMD,
drusens andAMD,
in the dry the white box
and the shows
white boxthe irregular
shows bloodblood
the irregular ves- vessel under
sel under the macula
theinmacula
the wetin AMD.
the wet AMD.

4.2. Model Training4.2. Model Training

All images were resized to 224 × 224 × 3 and divided into mini-batches. During the
All images were resized to 224 × 224 × 3 and divided into mini-batches. During the
process of retraining the ResNet-50, the Adam optimizer was used and the parameters of
process of retraining the ResNet-50,
all layers thewere
in ResNet-50 Adam optimizer
updated. was used
The batch and12,
size was thethe
parameters of was 0.001
learning rate
all layers in ResNet-50
and thewere
numberupdated. Thewas
of epochs batch
30. size was 12, the learning rate was 0.001
and the number of epochs was 30.
The study was implemented on a machine with an Intel (R) Xeon (R) CPU E5-2680
The study was
v4 @implemented on a machine
2.40 GHz processor with anTesla
and an NVIDIA IntelP40
(R)graphics
Xeon (R) CPU
card. TheE5-2680
experiments were
based on the open-source deep learning frameworks TensorFlow-gpu
v4 @ 2.40 GHz processor and an NVIDIA Tesla P40 graphics card. The experiments were 2.1.0 and Keras 2.3.1.
based on the open-source deep learning frameworks TensorFlow-gpu 2.1.0 and Keras
5. Results and Discussion
2.3.1.
5.1. Performance Evaluation
The results of the proposed work are evaluated based on the confusion matrix. In the
5. Results and Discussion
field of machine learning, the confusion matrix is a visual tool to measure the performance
5.1. Performance Evaluation
of classification. As shown in Table 2, each column of the matrix represents the label results
predicted by the work
The results of the proposed model,are
andevaluated
each row of the matrix
based on therepresents
confusionthe real label
matrix. of the sample.
In the
field of machine learning, the confusion matrix is a visual tool to measure the performance
Table 2. Binary-classification confusion matrix.
of classification. As shown in Table 2, each column of the matrix represents the label re-
sults predicted by the model, and each row of the matrix represents theLabel
Predict real label of the
True Label
sample. AMD NORMAL
AMD True Positive (TP) False Negative (FN)
Table 2. Binary-classification NORMAL
confusion matrix. False Positive (FP) True Negative (TN)

Predict Label
True Label In the confusion matrix, TP and TN give properly classified data results, while FP and
AMD NORMAL
FN give wrongly classified information. We can measure accuracy, precision, sensitivity,
AMD True Positive (TP) False Negative (FN)
and F1-score using these values to analyze the performance of the proposed method.
NORMAL Accuracy is the ratioFalse
ofPositive (FP)
the number True Negative
of samples classified correctly(TN)
to the total number
of OCT images, and is calculated using the following equation:
In the confusion matrix, TP and TN give properly classified data results, while FP
TP + TN
and FN give wrongly classified information. We can=measure accuracy, precision, sensi-
Accuracy (5)
TP + TN + FP + FN
tivity, and F1-score using these values to analyze the performance of the proposed
method.
Accuracy is the ratio of the number of samples classified correctly to the total number
of OCT images, and is calculated using the following equation:
 AMD, measuring the ability of the classifier to detect AMD images, and is calculated using
the following equation:

TP
Sensitivity = (6)
TP + FN
Diagnostics 2022, 12, 532 8 of 13
Precision represents the proportion of all images predicted to be AMD that are actu-
ally AMD images, and is calculated using the following equation:

TP images that are correctly classified as

Sensitivity refers to the proportion of all AMD
Pr ecision = detect AMD images, and is calculated(7)
AMD, measuring the ability of the classifier toTP + FP using
the following equation:
F1-score specifies the harmonic mean of precision TPand recall, and is calculated using
Sensitivity = (6)
the following equation: TP + FN
Precision represents the proportionProfecision × Sensitivit
all images predictedyto be AMD that are actually
AMD images, and is calculated = 2 × the following equation:
F1-scoreusing (8)
Pr ecision + Sensitivity
TP
the value=of AUC are used to measure the overall (7)
In addition, the ROC curve and Precision
TP + FP
classification performance of the proposed method.
F1-score specifies the harmonic mean of precision and recall, and is calculated using
[Link]
Results
following equation:
5.2.1. UCSD Dataset
Precision × Sensitivity
The confusion matrix ofF1the − score = 2×
proposed method, tested on the UCSD test set, is shown (8)
Precision + Sensitivity
in Figure 6a. The test set consisted of 250 dry AMD, 250 wet AMD, and 250 normal images,
so we tookIn addition, the AMD
a total of 500 ROC curve
imagesandandthe
250value of AUC
normal images arefor
used to measure
testing. All of the the
AMDoverall
classification
images performance
were correctly of the
classified, andproposed
in normalmethod.
images, only 1 image out of 250 was mis-
takenly classified as AMD. The ROC curve of the proposed approach on the UCSD test
set 5.2. Resultsin Figure 6c and the value of AUC is very close to 1.0000.
is plotted
5.2.1. UCSD
We also Dataset
directly used the retrained ResNet-50 as a classifier to verify the effectiveness
The confusion
of the combination matrix
of deep of the proposed
learning and LOF method, tested
algorithms. Theonconfusion
the UCSDmatrix
test set,
andis shown
the
ROC in curve
Figureare
[Link]
The test
inset consisted
Figure 6b,d, of 250 dry AMD,
respectively. 250 wet
In total, 1 outAMD, and
of 500 AMD250 normal
images images,
and
so of
4 out we250took a totalimages
normal of 500were
AMDclassified
images and 250 normal
mistakenly, and images
the value foroftesting.
the AUC Allwasof the
0.9998. The results show that deep learning combined with the LOF algorithm can classifywas
AMD images were correctly classified, and in normal images, only 1 image out of 250
OCT mistakenly
images more classified as AMD. The ROC curve of the proposed approach on the UCSD test
efficiently.
set is plotted in Figure 6c and the value of AUC is very close to 1.0000.

Diagnostics 2022, 12, x FOR PEER REVIEW 9 of 14

(a) (b)

(c) (d)
Figure
Figure6.6.(a)
(a)Confusion
Confusionmatrix
matrixofofthe
theproposed
proposedmethod
methodon onthe
theUCSD
UCSDtest
testset.
set.(b)
(b)Confusion
Confusionmatrix
matrix
ofofResNet-50
ResNet-50ononthe UCSD
the UCSDtest [Link].
test (c) (c)
ROC ROCcurve of the
curve proposed
of the method
proposed on the
method onUCSD test set.
the UCSD test(d)
set.
ROC curve of ResNet-50 on the UCSD test set.
(d) ROC curve of ResNet-50 on the UCSD test set.

The accuracy, sensitivity, precision, F1-score and AUC obtained on the UCSD test set
are shown in Table 3. It is clear that the AMD possessed a better accuracy than normal.
The weighted average evaluation parameters, such as F1-score, sensitivity, and precision,
were greater than 99.80%, with an AUC value very close to 1.0000. Hence, it shows that
the proposed method performs well in the classification of AMD and normal OCT images.
Diagnostics 2022, 12, 532 9 of 13

We also directly used the retrained ResNet-50 as a classifier to verify the effectiveness
of the combination of deep learning and LOF algorithms. The confusion matrix and the
ROC curve are shown in Figure 6b,d, respectively. In total, 1 out of 500 AMD images and 4
out of 250 normal images were classified mistakenly, and the value of the AUC was 0.9998.
The results show that deep learning combined with the LOF algorithm can classify OCT
images more efficiently.
The accuracy, sensitivity, precision, F1-score and AUC obtained on the UCSD test set
are shown in Table 3. It is clear that the AMD possessed a better accuracy than normal.
The weighted average evaluation parameters, such as F1-score, sensitivity, and precision,
were greater than 99.80%, with an AUC value very close to 1.0000. Hence, it shows that the
proposed method performs well in the classification of AMD and normal OCT images.

Table 3. Quantitative results of the proposed method on the UCSD test set.

Class Accuracy % Sensitivity % Precision % F1-Score % AUC

AMD 100.00 100.00 99.80 99.90 -
NORMAL 99.60 99.60 100.00 99.80 -
Weighted Average 99.87 99.87 99.87 99.87 1.0000

A comparison of the proposed method with existing methods conducted on the UCSD
dataset is given in Table 4. Thomas et al. [18,19] proposed two novel CNN architectures
specifically for AMD and normal classification based on OCT images, obtaining an accuracy
of 99.78% and 99.73%, with an AUC of 0.9978 and 0.9999, respectively. Other previous
works obtained less than 98.50% accuracy and less than 0.9920 AUC. In the case of the
UCSD dataset, the proposed method obtained a higher average accuracy and AUC than
other methods and showed better performance in detecting AMD.

Table 4. Comparison of the proposed method with existing works conducted on the UCSD dataset.

Method Weighted Average Accuracy % AUC

Multi-scale and multi-path CNN [31] 99.78 0.9978
Multi-scale CNN [32] 99.73 0.9999
Inception V3 transfer learning [42] 96.53 0.9762
Multi-scale deep feature fusion [44] 97.71 0.9900
AlexNet transfer learning [45] 98.26 0.9917
Iterative fusion CNN [46] 93.40 0.9798
Proposed method 99.87 1.0000

5.2.2. Duke Dataset

The proposed method was trained only on the UCSD dataset but was tested on two
publicly available datasets, the UCSD dataset and the Duke dataset. The Duke dataset
consists of 1407 normal images, and 723 AMD images [30]. The confusion matrix of the
Duke test set is given in Figure 7a. Out of 723 AMD images, 36 of them were misclassified
as normal. In normal images, 16 images out of 1407 were misclassified as AMD. The ROC
curve of the proposed work on the Duke test set is plotted in Figure 7c, and the value of
AUC was approximately 0.9954. The confusion matrix and the ROC curve when testing
with ResNet-50 on the Duke test set are also given in Figure 7b,d, respectively. It is obvious
that the proposed method, combining ResNet-50 and the LOF algorithm, performed better
than using ResNet-50 alone.
The performance evaluation of the Duke test set is tabulated in Table 5. The results of
the proposed method tested on the Duke dataset are not as excellent as those in the UCSD
dataset, but still obtained an overall weighted accuracy of 97.56%.
 curve of the proposed work on the Duke test set is plotted in Figure 7c, and the value of
AUC was approximately 0.9954. The confusion matrix and the ROC curve when testing
with ResNet-50 on the Duke test set are also given in Figure 7b,d, respectively. It is obvi-
ous that the proposed method, combining ResNet-50 and the LOF algorithm, performed
Diagnostics 2022, 12, 532 10 of 13
better than using ResNet-50 alone.

(a) (b)

(c) (d)
Figure 7. (a)Figure
Confusion matrix of the
7. (a) Confusion proposed
matrix method method
of the proposed on the Duke
on thetest
[Link]
(b)set.
Confusion matrix
(b) Confusion matrix of
of ResNet-50 on the Duke test set. (c) ROC curve of the proposed method on the Duke test
ResNet-50 on the Duke test set. (c) ROC curve of the proposed method on the Duke test set. set. (d)(d) ROC
ROC curve curve
of ResNet-50 on the
of ResNet-50 onDuke test set.
the Duke test set.

The performance evaluation

Table 5. Quantitative resultsofonthe
theDuke testset.
Duke test set is tabulated in Table 5. The results
of the proposed method tested on the Duke dataset are not as excellent as those in the
UCSD dataset, but Class Accuracy
still obtained %
an overall Sensitivity accuracy
weighted % Precision %
of 97.56%. F1-Score % AUC
The comparisonAMD of the proposed95.02 method 95.02 with existing 97.72
methods conducted
96.38 on the-
Duke dataset isNORMAL
given in Table 6.98.86
The approach98.86 proposed by 97.48
Hussain et al.98.17
[35] obtained-
the best classification accuracy of97.56
Weighted Average 97.70%, with97.56an AUC of 0.9900.
97.56 First, they
97.56segmented 0.9954
the retina layers, then extracted features such as the thickness of the retina and the thick-
ness of the individual retinal layers,
The comparison of the and the volume
proposed of the
method pathologies
with such as drusen
existing methods conductedandon the
hyper-reflective intra-retinal
Duke dataset spots.
is given Finally,
in Table theapproach
6. The classification resultby
proposed was obtained
Hussain using
et al. [35] an
obtained
RF classifier. The other previous works achieved an accuracy of less than 97.00%.
the best classification accuracy of 97.70%, with an AUC of 0.9900. First, they segmented the
retina layers, then extracted features such as the thickness of the retina and the thickness
Table 5. Quantitative results onretinal
of the individual the Duke test set.
layers, and the volume of the pathologies such as drusen and
hyper-reflective intra-retinal spots. Finally, the classification result was obtained using an
Class Accuracy % Sensitivity % Precision % F1-Score % AUC
RF classifier. The other previous works achieved an accuracy of less than 97.00%.
AMD 95.02 95.02 97.72 96.38 -
NORMAL 98.86 Table 6. 98.86 97.48 work with existing
Comparison of the proposed 98.17 - Duke dataset.
works conducted on the
ighted Average 97.56 97.56 97.56 97.56 0.9954
Method Weighted Average Accuracy % AUC
RPE detection and baseline estimation [26] 96.66 -
Feature extraction + SVM [43] 93.30 -
Intensity-based threshold + Ploy fitting curve [47] 92.00 -
Feature extraction + RF classifier [48] 97.70 0.9900
Feature extraction + Sequential Minimal
96.60 0.9910
Optimization [49]
18-layer recombined residual CNN [50] 96.66 -
Sparse coding +Dictionary learning [51] 96.66 -
Proposed method 97.56 0.9954
Diagnostics 2022, 12, 532 11 of 13

According to the above analysis, even if the proposed method is trained only on the
UCSD dataset, it also achieved good detection accuracy when tested on the Duke dataset.
Additionally, the proposed method was compared with the method using ResNet-50 alone,
as shown in Figures 6 and 7. The comparison was also made with previous methods, such
as those in [26,31,32,42–51], as shown in Tables 4 and 6, based on accuracy and AUC. The
results suggest that the proposed method performed very well compared with previous
models when classifying AMD and normal OCT images.

6. Conclusions
A novel automatic detection method was presented for the detection of AMD from
OCT images based on deep learning and an outlier detection method. The ResNet-50
model with L2-constrained softmax loss was retrained to extract features from OCT images,
and the LOF algorithm was used as the classifier. The proposed method was trained on
the UCSD dataset and tested on both the UCSD dataset and the Duke dataset, with an
accuracy of 99.87% and 97.56%, respectively. Even though the model was only trained on
the UCSD dataset, it also performed well when tested on the Duke dataset. Tables 4 and 6
show the comparison of the proposed method with existing works, which also indicates
the efficiency of the proposed method in detecting AMD.
The advantage of the proposed method is its excellent ability to classify AMD and
normal OCT images without preprocessing and with high accuracy, which will help doctors
in large-scale OCT image screening.
Our proposed method achieves a good overall performance in the detection of AMD,
enabling the proposed method to be used for the early detection of AMD. In future work,
we hope to use more datasets to validate the proposed method, and based on this study,
we will further subdivide the AMD into dry AMD and wet AMD.

Author Contributions: Project administration, Y.Z.; methodology, T.H. and Q.Z.; software, T.H.;
supervision, Y.Z.; writing—original draft preparation, T.H.; writing—review and editing, Q.Z. and
Y.Z. All authors have read and agreed to the published version of the manuscript.
Funding: This research received no external funding.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: No new dataset was generated from this study. We utilized the follow-
ing two public datasets in this study: [Link] (accessed on
19 January 2022) and [Link] (accessed
on 19 January 2022).
Acknowledgments: This research did not receive any specific grant from funding agencies in the
public, commercial, or not-for-profit sectors.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. GBD 2017 Disease and Injury Incidence and Prevalence Collaborators. Global, regional, and national incidence, prevalence, and
years lived with disability for 354 diseases and injuries for 195 countries and territories, 1990–2017: A systematic analysis for the
Global Burden of Disease Study 2017. Lancet 2018, 392, 1789–1858. [CrossRef]
2. Heesterbeek, T.J.; Lorés-Motta, L.; Hoyng, C.B.; Lechanteur, Y.; den Hollander, A.I. Risk factors for progression of age-related
macular degeneration. Ophthalmic Physiol. Opt. 2020, 40, 140–170. [CrossRef] [PubMed]
3. Wong, W.L.; Su, X.; Li, X.; Cheung, C.M.; Klein, R.; Cheng, C.Y.; Wong, T.Y. Global prevalence of age-related macular degeneration
and disease burden projection for 2020 and 2040: A systematic review and meta-analysis. Lancet Glob. Health 2014, 2, 106–116.
[CrossRef]
4. Schultz, N.M.; Bhardwaj, S.; Barclay, C.; Gaspar, L.; Schwartz, J. Global Burden of Dry Age-Related Macular Degeneration: A
Targeted Literature Review. Clin. Ther. 2021, 43, 1792–1818. [CrossRef] [PubMed]
5. Stahl, A. The Diagnosis and Treatment of Age-Related Macular Degeneration. Dtsch. Arztebl. Int. 2020, 117, 513–520. [CrossRef]
[PubMed]
Diagnostics 2022, 12, 532 12 of 13

6. Davis, M.D.; Gangnon, R.E.; Lee, L.Y.; Hubbard, L.D.; Klein, B.E.; Klein, R.; Ferris, F.L.; Bressler, S.B.; Milton, R.C. The Age-Related
Eye Disease Study severity scale for age-related macular degeneration: AREDS Report No. 17. Arch. Ophthalmol. 2005, 123,
1484–1498. [CrossRef]
7. Gheorghe, A.; Mahdi, L.; Musat, O. Age-related macular degeneration. Rom. J. Ophthalmol. 2015, 59, 74–77.
8. Mitchell, P.; Liew, G.; Gopinath, B.; Wong, T.Y. Age-related macular degeneration. Lancet 2018, 392, 1147–1159. [CrossRef]
9. Huang, D.; Swanson, E.A.; Lin, C.P.; Schuman, J.S.; Stinson, W.G.; Chang, W.; Hee, M.R.; Flotte, T.; Gregory, K.; Puliafito, C.A.
Optical coherence tomography. Science 1991, 254, 1178–1181. [CrossRef]
10. Swanson, E.A.; Izatt, J.A.; Hee, M.R.; Huang, D.; Lin, C.P.; Schuman, J.S.; Puliafito, C.A.; Fujimoto, J.G. In vivo retinal imaging by
optical coherence tomography. Opt. Lett. 1993, 18, 1864–1866. [CrossRef]
11. Balasubramani, V.; Kujawińska, M.; Allier, C.; Anand, V.; Cheng, C.J.; Depeursinge, C.; Hai, N.; Juodkazis, S.; Kalkman, J.; Kuś, A.;
et al. Roadmap on Digital Holography-Based Quantitative Phase Imaging. J. Imaging 2021, 7, 252. [CrossRef] [PubMed]
12. Rosen, J.; Alford, S.; Anand, V.; Art, J.; Bouchal, P.; Bouchal, Z.; Erdenebat, M.U.; Huang, L.; Ishii, A.; Juodkazis, S.; et al. Roadmap
on Recent Progress in FINCH Technology. J. Imaging 2021, 7, 197. [CrossRef] [PubMed]
13. Puliafito, C.A.; Hee, M.R.; Lin, C.P.; Reichel, E.; Schuman, J.S.; Duker, J.S.; Izatt, J.A.; Swanson, E.A.; Fujimoto, J.G. Imaging of
macular diseases with optical coherence tomography. Ophthalmology 1995, 102, 217–229. [CrossRef]
14. Araveti, S.K.; Hiraishi, N.; Kominami, N.; Otsuki, M.; Sumi, Y.; Yiu, C.; Tagami, J. Swept-source optical coherence tomographic
observation on prevalence and variations of cemento-enamel junction morphology. Lasers Med. Sci. 2020, 35, 213–219. [CrossRef]
15. Yow, A.P.; Srivastava, R.; Cheng, J.; Li, A.; Liu, J.; Schmetterer, L.; Tey, H.L.; Wong, D. Techniques and Applications in Skin OCT
Analysis. Adv. Exp. Med. Biol. 2020, 1213, 149–163. [CrossRef]
16. Wang, J.; Paritala, P.K.; Mendieta, J.B.; Komori, Y.; Raffel, O.C.; Gu, Y.; Li, Z. Optical coherence tomography-based patient-
specific coronary artery reconstruction and fluid-structure interaction simulation. Biomech. Modeling Mechanobiol. 2020, 19, 7–20.
[CrossRef]
17. Wong, P.W.; Guo, J.; Khanzada, N.K.; Yim, V.; Kyoungjin, A. In-situ 3D fouling visualization of membrane distillation treating
industrial textile wastewater by optical coherence tomography imaging. Water Res. 2021, 205, 117668. [CrossRef]
18. Larimer, C.J.; Denis, E.H.; Suter, J.D.; Moran, J.J. Optical coherence tomography imaging of plant root growth in soil. Appl. Opt.
2020, 59, 2474–2481. [CrossRef]
19. Elsharkawy, M.; Elrazzaz, M.; Ghazal, M.; Alhalabi, M.; Soliman, A.; Mahmoud, A.; El-Daydamony, E.; Atwan, A.; Thanos, A.;
Sandhu, H.S.; et al. Role of Optical Coherence Tomography Imaging in Predicting Progression of Age-Related Macular Disease: A
Survey. Diagnostics 2021, 11, 2313. [CrossRef]
20. Costa, R.A.; Skaf, M.; Melo, L.A., Jr.; Calucci, D.; Cardillo, J.A.; Castro, J.C.; Huang, D.; Wojtkowski, M. Retinal assessment using
optical coherence tomography. Prog. Retin. Eye Res. 2006, 25, 325–353. [CrossRef]
21. van Velthoven, M.E.; Faber, D.J.; Verbraak, F.D.; van Leeuwen, T.G.; de Smet, M.D. Recent developments in optical coherence
tomography for imaging the retina. Prog. Retin. Eye Res. 2007, 26, 57–77. [CrossRef] [PubMed]
22. Regatieri, C.V.; Branchini, L.; Duker, J.S. The role of spectral-domain OCT in the diagnosis and management of neovascular
age-related macular degeneration. Ophthalmic Surg. Lasers Imaging 2011, 42, 56–66. [CrossRef] [PubMed]
23. Farsiu, S.; Chiu, S.J.; O’Connell, R.V.; Folgar, F.A.; Yuan, E.; Izatt, J.A.; Toth, C.A.; Age-Related Eye Disease Study 2 Ancillary
Spectral Domain Optical Coherence Tomography Study Group. Quantitative classification of eyes with and without intermediate
age-related macular degeneration using optical coherence tomography. Ophthalmology 2014, 121, 162–172. [CrossRef] [PubMed]
24. Naz, A.; Ahmed, A.; Akram, M.U.; Khan, S.A. Automated segmentation of RPE layer for the detection of age macular degeneration
using OCT images. In Proceedings of the 2016 Sixth International Conference on Image Processing Theory, Tools and Applications,
Oulu, Finland, 12–15 December 2016. [CrossRef]
25. Arabi, P.M.; Krishna, N.; Ashwini, V.; Prathibha, H.M. Identification of Age-Related Macular Degeneration Using OCT Images.
Int. Conf. Adv. Mater. Manuf. Appl. 2018, 310, 012096. [CrossRef]
26. Thomas, A.; Sunija, A.P.; Manoj, R.; Ramachandran, R.; Ramachandran, S.; Varun, P.G.; Palanisamy, P. RPE layer detection and
baseline estimation using statistical methods and randomization for classification of AMD from retinal OCT. Comput. Methods
Programs Biomed. 2021, 200, 105822. [CrossRef]
27. Sharif, M.M.; Akram, M.U.; Malik, A.W. Extraction and Analysis of RPE layer from OCT Images for Detection of Age Related
Macular Degeneration. In Proceedings of the 2018 IEEE 20th International Conference on e-Health Networking, Applications and
Services (Healthcom), Ostrava, Czech Republic, 17–20 September 2018. [CrossRef]
28. Gu, J.; Wang, Z.; Kuen, J.; Ma, L.; Shahroudy, A.; Shuai, B.; Liu, T.; Wang, X.; Wang, G.; Cai, J.; et al. Recent advances in
convolutional neural networks. Pattern Recognit. 2018, 77, 354–377. [CrossRef]
29. Lee, C.S.; Baughman, D.M.; Lee, A.Y. Deep learning is effective for the classification of OCT images of normal versus Age-related
Macular Degeneration. Ophthalmol. Retin. 2017, 1, 322–327. [CrossRef]
30. Serener, A.; Serte, S. Dry and Wet Age-Related Macular Degeneration Classification Using OCT Images and Deep Learning. In
Proceedings of the 2019 Scientific Meeting on Electrical-Electronics & Biomedical Engineering and Computer Science (EBBT),
Istanbul, Turkey, 24–26 April 2019. [CrossRef]
31. Thomas, A.; Harikrishnan, P.M.; Ramachandran, R.; Ramachandran, S.; Manoj, R.; Palanisamy, P.; Gopi, V.P. A novel multiscale
and multipath convolutional neural network based age-related macular degeneration detection using OCT images. Comput.
Methods Programs Biomed. 2021, 209, 106294. [CrossRef]
Diagnostics 2022, 12, 532 13 of 13

32. Thomas, A.; Harikrishnan, P.M.; Krishna, A.K.; Palanisamy, P.; Gopi, V.P. A novel multiscale convolutional neural network based
age-related macular degeneration detection using OCT images. Biomed. Signal Process. Control 2021, 67, 102538. [CrossRef]
33. Yoo, T.K.; Choi, J.Y.; Seo, J.G.; Ramasubramanian, B.; Selvaperumal, S.; Kim, D.W. The possibility of the combination of OCT
and fundus images for improving the diagnostic accuracy of deep learning for age-related macular degeneration: A preliminary
experiment. Med. Biol. Eng. Comput. 2019, 57, 677–687. [CrossRef]
34. Kadry, S.; Rajinikanth, V.; González Crespo, R.; Verdú, E. Automated detection of age-related macular degeneration using a
pre-trained deep-learning scheme. J. Supercomput. 2021. [CrossRef]
35. He, K.; Zhang, X.; Ren, S.; Sun, J. Deep residual learning for image recognition. In Proceedings of the 2016 IEEE Conference on
Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA, 26 June–1 July 2016; pp. 770–778. [CrossRef]
36. Lu, J.; Behbood, V.; Hao, P.; Zuo, H.; Xue, S.; Zhang, G. Transfer learning using computational intelligence: A survey. Knowl.-Based
Syst. 2015, 80, 14–23. [CrossRef]
37. Breunig, M.M.; Kriegel, H.; Ng, R.T.; Sander, J. LOF: Identifying density-based local outliers. ACM SIGMOD Rec. 2000, 29, 93–104.
[CrossRef]
38. Baltruschat, I.M.; Nickisch, H.; Grass, M.; Knopp, T.; Saalbach, A. Comparison of Deep Learning Approaches for Multi-Label
Chest X-Ray Classification. Sci. Rep. 2019, 9, 6381. [CrossRef]
39. Serte, S.; Demirel, H. Deep learning for diagnosis of COVID-19 using 3D CT scans. Comput. Biol. Med. 2021, 132, 104306.
[CrossRef]
40. Khojasteh, P.; Passos Júnior, L.A.; Carvalho, T.; Rezende, E.; Aliahmad, B.; Papa, J.P.; Kumar, D.K. Exudate detection in fundus
images using deeply-learnable features. Comput. Biol. Med. 2019, 104, 62–69. [CrossRef]
41. Ranjan, R.; Castillo, C.D.; Chellappa, R. L2-constrained Softmax Loss for Discriminative Face Verification. arXiv 2017,
arXiv:1703.09507.
42. Kermany, D.S.; Goldbaum, M.; Cai, W.; Valentim, C.; Liang, H.; Baxter, S.L.; McKeown, A.; Yang, G.; Wu, X.; Yan, F.; et al.
Identifying Medical Diagnoses and Treatable Diseases by Image-Based Deep Learning. Cell 2018, 172, 1122–1131. [CrossRef]
43. Srinivasan, P.P.; Kim, L.A.; Mettu, P.S.; Cousins, S.W.; Comer, G.M.; Izatt, J.A.; Farsiu, S. Fully automated detection of diabetic
macular edema and dry age-related macular degeneration from optical coherence tomography images. Biomed. Opt. Express 2014,
5, 3568–3577. [CrossRef]
44. Das, V.; Dandapat, S.; Bora, P.K. Multi-scale deep feature fusion for automated classification of macular pathologies from OCT
images. Biomed. Signal Process. Control 2019, 54, 101605. [CrossRef]
45. Kaymak, S.; Serener, A. Automated Age-Related Macular Degeneration and Diabetic Macular Edema Detection on OCT Images
using Deep Learning. In Proceedings of the 2018 IEEE 14th International Conference on Intelligent Computer Communication
and Processing (ICCP), Cluj-Napoca, Romania, 6–8 September 2018; pp. 265–269. [CrossRef]
46. Fang, L.; Jin, Y.; Huang, L.; Guo, S.; Zhao, G.; Chen, X. Iterative fusion convolutional neural networks for classification of optical
coherence tomography images. J. Vis. Commun. Image Represent. 2019, 59, 327–333. [CrossRef]
47. Khalid, S.; Akram, M.U.; Jameel, A.; Khalil, T. Automated Detection of Drusens to Diagnose Age Related Macular Degeneration
Using OCT Images. IJCSIS 2016, 14. Available online: [Link]
Detection_of_Drusens_to_Diagnose_Age_Related_Macular_Degeneration_Using_OCT_Images (accessed on 19 January 2022).
48. Hussain, M.A.; Bhuiyan, A.; Luu, C.D.; Theodore Smith, R.; Guymer, R.H.; Ishikawa, H.; Schuman, J.S.; Ramamohanarao, K.
Classification of healthy and diseased retina using SD-OCT imaging and Random Forest algorithm. PLoS ONE 2018, 13, 0198281.
[CrossRef] [PubMed]
49. Wang, Y.; Zhang, Y.; Yao, Z.; Zhao, R.; Zhou, F. Machine learning based detection of age-related macular degeneration (AMD)
and diabetic macular edema (DME) from optical coherence tomography (OCT) images. Biomed. Opt. Express 2016, 7, 4928–4940.
[CrossRef] [PubMed]
50. Meng, T.; Wu, C.; Jia, T.; Jiang, Y.; Jia, Z. Recombined Convolutional Neural Network for Recognition of Macular Disorders
in SD-OCT Images. In Proceedings of the 2018 37th Chinese Control Conference (CCC), Wuhan, China, 25–27 July 2018; pp.
9362–9367. [CrossRef]
51. Sun, Y.; Li, S.; Sun, Z. Fully automated macular pathology detection in retina optical coherence tomography images using sparse
coding and dictionary learning. J. Biomed. Opt. 2017, 22, 16012. [CrossRef]