You are on page 1of 9

Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Automatic Detection of Knee Joints and


Quantification of Knee Osteoarthritis Severity
using Modified Fully connected Convolutional
Neural Networks
1
D.Kiruthika, 2J. Judith
1
Junior Research fellow, ECE, 2Assistant professor, ECE,
Sethu Institute of Technology, Kariapatti, Virudhunagar, Tamilnadu, India

Abstract:- Knee Osteoarthritis (OA) is an extremely To automatically localise the knee joints, we present a
common and degenerative musculoskeletal disease fully-convolutional neural network (FCN)-based method. A
worldwide which creates a significant burden on patients FCN is an end-to-end network that has been trained to make
with reduced quality of life and also on society because pixel-by-pixel predictions [3]. To automatically classify the
of its financial impact. Therefore, technical try and localized knee joints, we propose two methods: 1) training a
efforts to reduce the burden of the disease could help CNN from scratch for multi-class classification of knee OA
both patients and society. In this paper, an automated images, and 2) training a CNN to optimize a weighted ratio
novel method is proposed with a supported combination of two loss functions: categorical cross-entropy for multi-
of joint shape and modified Fully connected neural class classification and mean-squared error for
network (FCNN) based bone texture features, to reclassification. We compare the results of these methods to
differentiate between the knee radiographs with and those of WND-CHARM [4] and our previous research. We
without osteoarthritis. Moreover, an endeavor is formed also compare the classification results to both manual and
to explain the bone texture using CNN. Knee automated methods.We propose a novel pipeline for
radiographs from Osteoarthritis Initiative (OAI) and automatically quantifying knee OA severity, which
Multicenter Osteoarthritis (MOST) datasets are utilized includes an FCN for localising knee joints and a CNN
in this paper. The proposed models were trained on 8000 jointly trained for knee joint classification and
knee radiographs from OAI and evaluated on 3500 knee regression. The main contributions of this work
radiographs from MOST. The results demonstrate that include a fully-convolutional network (FCN)-based
fusing the proposed shape and texture parameters method for automatically localising knee joints, as
achieves the state-of-the art performance in well as training a network (CNN) from scratch that
radiographic OA detection yielding area under the ROC optimises a weighted ratio of both categorical cross-
curve (AUC) of 98.75% accuracy. entropy and mean-squared error for knee joint
regression. This multi-objective convolutional learning
Keywords:- Knee Osteoarthritis, KL grades, Automatic improves overall quantification while also providing
Detection, Fully Convolutional Neural Networks, multi-class classification and regression outputs at the
Classification and Regression. same time.
I. INTRODUCTION II. LITERATURE SURVEY
Knee Osteoarthritis (OA) may be a debilitating joint Classifying the severity of knee OA can be
disorder that mainly degrades the knee articular cartilage. accomplished by detecting variations in joint space width
Clinically, the most important pathological features of knee and the formation of osteophytes in the knee joints [5]. Yoo
OA include joint space narrowing, osteophyte formation, et al., in a recent approach, developed a scoring system to
and sclerosis. Knee OA has a high incidence among the predict radiographic and symptomatic knee OA risks using
elderly, obese, and people with a sedentary lifestyle. In its artificial neural networks (ANN) and KNHANES V-1 data.
severe stages, it causes excruciating pain and ends up in Then, the Shamir et al., proposed a WND-CHARM method ,
total joint arthroplasty. In the Early diagnosis period , it is a which is a multipurpose bio-medical image classifier [6]
crucial for clinical treatments and pathology [1]. Despite the was used to classify knee OA radiographs [7] and to detect
introduction of several imaging modalities like, Optical knee OA early using computer aided analysis. WND-
Coherence Tomography,MRI and radiography (X-ray) , an CHARM extracts handcrafted features from raw images and
ultrasound for augmented OA diagnosis, has been image transforms [8].
traditionally preferredand “gold standard” for preliminary
diagnosis of knee OA [2]. During this work, we train CNNs Convolutional neural networks (CNNs) have
from scratch to automatically quantify knee OA severity recently outperformed many methods based on hand-
using X-ray images. This involves two main steps: 1) crafted features in many computer vision tasks such as
automatically detecting and extracting the region of interest image recognition, automatic detection and segmentation,
(ROI) and localizing the knee joints, and 2) classifying the content-based image retrieval, and video classification.
localized knee joints. CNNs learn effective feature representations that are

IJISRT22DEC670 www.ijisrt.com 569


Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
particularly well-suited for fine-grained classification [9], III. MATERIALS AND METHODS
such as knee OA image classification. In a previous
study, we demonstrated that off-the-shelf CNNs trained A. Data Acquisition
on the ImageNet LSVRC dataset [13], such as the VGG The data used in this study for the experiments and
16-Layers network [10], the VGG-M-128 network [11], analysis are bilateral PA fixed flexion knee X-ray
and the BVLC reference CaffeNet [12], can be fine-tuned images. The datasets are from the University of
for classifying knee OA images using transfer learning. California, San Francisco's Osteoarthritis Initiative
We also argued that instead of binary or multi-class (OAI) and Multicenter Osteoarthritis Study (MOST),
classification, it is preferable to assess knee OA severity and are standard datasets used in knee osteoarthritis
using a continuous metric such as mean-squared error studies.
and showed that predicting the continuous grades through
regression reduces the mean-squared error and in turn B. Kellgren and Lawrence Grades
improves the overall quantification. Before that, Shamir et The Kellgren and Lawrence (KL) grades are used as the
al. et al. [14] proposed using template matching to detect ground truth in this study to classify the knee OA X-ray
and extract knee joints automatically. For large datasets like images. The KL grading system is still regarded as the gold
OAI, this method is slow, and the accuracy and precision of standard for assessing the severity of knee osteoarthritis in
detecting knee joints is low. In a previous study, we radiographs [16]. It assigns a five-point scale to the severity
presented an SVM-based method for automatically detecting of radiographic knee OA. 'Grade 0' means normal, 'Grade 1'
the centre of knee joints [15] and extracting a fixed region as means doubtful, 'Grade 2' means minimal, 'Grade 3' means
the ROI with reference to the detected centre. This method moderate, and 'Grade 4' means severe. The KL grading
is also not very accurate, and the aspect ratio of the system is depicted in Figure 1.
extracted knee joints is compromised, which affects the
overall quantification. C. OAI and MOST Datasets
The OAI dataset's baseline cohort includes MRI and X-
ray images from 4,476 participants. We chose 4,446 X-ray
images from the entire cohort based on the availability of
KL grades for both knees as determined by the Boston
University X-ray reading centre (BU). There are 8,892 knee
images in total, with the following distribution based on KL
grades: Grade 0 - 3433, Grade 1 - 1589, Grade 2 - 2353,
Grade 3 - 1222, and Grade 4 - 295.The MOST dataset
contains lateral knee radiograph assessments from 3,026
people. Based on the availability of KL grades for both
knees as per baseline to 84-month Longitudinal Knee
Radiograph Assessments, 2,920 radiographs are chosen.
This dataset contains 5,840 knee images and the distribution
as per KL grades is as follows: Grade 0 - 2497, Grade 1 -
1017, Grade 2 - 922, Grade 3 - 970, and Grade 4 - 431.

Fig. 1: The KL grading system to assess the severity of knee OA [16]

IJISRT22DEC670 www.ijisrt.com 570


Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IV. PROPOSED METHODOLOGY methods for automatically localizing knee joints, such as
template matching [18] and our own SVM-based method,
This section describes the methodology for calculating were ineffective. In this paper, we propose a fully
radiographic knee OA severity. This consists of two steps: convolutional neural network (FCN)-based method for
automatically detecting knee joints with a fully improving the accuracy and precision of detecting knee
convolutional network (FCN) and simultaneously joints.
classifying and localized knee images with a convolutional
neural network (CNN). Figure 2 depicts the entire pipeline B. Fully connected Convolutional Neural Network
used to assess the severity of knee OA. Architecture
We trained a fully convolutional neural network
A. Automatically Localizing Knee Joints using a FCN (FCN) to automatically detect the region of interest
architecture (ROI): the knee joints from knee OA radiographs,
The severity of knee OA can be determined by inspired by the success of a fully convolutional neural
detecting variations in joint space width and the network (FCN) for semantic segmentation on general
formation of osteophytes in the knee joint [17]. Thus, images. Our proposed FCN is built on a lightweight
localizing the knee joints from X-ray images is an architecture, and the network parameters are learned
important pre-processing step before quantifying knee from the scratch. The architecture is depicted in Figure
OA severity, and automatic methods are preferable for 2. After the experiment analysis, we discovered that
larger datasets. Figure 3 depicts a knee OA radiograph this architecture is the best for detecting knee joints.
and the region of interest (ROI) for detection. Previous

Fig. 2: The proposed architecture for detecting Knee Osteoarthritis.

The network is composed of four stages of The network's input is [256 256], and its output is the
convolutions, each followed by a max-pooling layer, same size.
and the final stage of convolutions is followed by an
up-sampling and a fully-convolutional layer. The first C. FCN Training
and second layer of convolution employ 32 filters The network was trained from scratch using training
each, the third layer consist of 64 filters, and the fourth samples of knee OA radiographs from the OAI and MOST
layer with 96 filters. The network employs uniform [3 datasets. Binary images with masks specifying the ROI: the
3] convolution and max pooling [2 2]. Following each knee joints are used to train the network. Figure 4 depicts an
convolution layer is a batch normalization and a example of a binary mask: the ground truth. The binary
rectified linear unit activation layer (ReLU). As the masks were generated from manual annotations of knee OA
network employs three stages of [2 2] max pooling, a radiographs using a fast annotation tool that we created. The
[8 8] up-sampling is performed after the final network was trained to minimise the total binary cross
convolution layer. Up-sampling is required for end-to- entropy between predicted and true pixels. We used the
end learning via back propagation from pixel-wise loss adaptive moment estimation (Adam) optimizer [20] with
and to obtain pixel-dense results. For pixel-based default parameters and discovered that it provided faster
classification, the final layer is a fully convolutional convergence than standard SGD. Figure 3 shows that the
layer with a kernel size of [1 1] and sigmoid activation. ROI (Region of Interest) regions are marked and segmented.

IJISRT22DEC670 www.ijisrt.com 571


Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Fig. 3: Selection of ROI in knee X-ray image

Fig. 4: The Fully Convolutional Network for automatically detecting knee joints

In the figure 4, the knee joints are extracted and Following the final pooling layer is a fully
deduce the bounding boxes of the knee joints using connected layer and a softmax dense layer. We
simple contour detection from the output predictions of include a drop out layer to avoid over-fitting. In the
FCN. n the proposed architecture, four convolutional final two convolutional layers (conv3 and conv4), as
layers are added , then four Max-pooling layers are added well as the fully connected layer, we apply an L2-norm
to reduce the dimensionality of the datasets. The fully weight regularization penalty of 0.01. (fc5). Applying a
connected layers are added at the end of the architecture. regularization penalty to other layers lengthens training
The output of the max-pooling layers is given as an input time while introducing no significant variation in
to the fully connected layers. The output from this layers learning curves. Using the Adam optimizer, the
are flattened, the weights are compared with the flattened network was trained to minimize categorical cross-
value to predict the Normal and OA images, and extract entropy loss [20]. The network's inputs are knee images
the knee joints from knee OA radiographs using the with dimensions of [200 300]. Based on the mean
bounding boxes. We upscale the bounding boxes from aspect ratio (1.6) of all extracted knee joints, we chose
the FCN output of size [256 256] to the original size of this size to roughly preserve the aspect ratio.
each knee OA radiograph before extracting the knee
joints to preserve the aspect ratio of the knee joints. E. Jointly training CNN for Classification and Regression
In general, assessing the severity of knee OA is
D. Quantifying knee OA severity using Convolutional based on multi-classifying knee images and assigning a
Neural Network KL grade to each distinct category [10]. We argued in
Convolutional neural networks are trained from our previous paper [1] that assigning a continuous
scratch using knee OA data, and networks are jointly grade (0-4) to knee images through regression is a
trained to minimize classification and regression losses, better approach for quantifying knee OA severity
allowing for a more accurate assessment of knee OA because the disease is progressive. However, there is no
severity. ground truth of KL grades on a continuous scale to
train a network directly for regression output with this
 Training CNN for Classification: The network is approach. As a result, we use multi-objective
made up of five learned weight layers: four convolutional learning [8] to train networks to optimize
convolutional layers and one fully connected layer. a weighted-ratio of two loss functions: categorical
The network architecture is depicted in Figure 5. cross-entropy and mean-squared error. Mean squared
Due to the scarcity of training data, we considered a error informs the network about grade ordering, while
lightweight architecture with few layers, and the cross entropy informs it about grade quantization.
network has 5.4 million free parameters in total. We Optimizing a network with two loss functions
discovered that this architecture is the best for intuitively provides a stronger error signal and is a step
classifying knee images after experimenting with toward improving overall quantification, taking both
the number of convolutional layers and other classification and regression results into account. We
parameters. The network's convolutional layers are arrived at the final architecture shown in Figure6 after
followed by batch normalization and a rectified much experimentation. This network has six learned
linear unit activation layer (ReLU). There is a max weight layers: five convolutional layers and one fully
pooling layer after each convolutional stage.

IJISRT22DEC670 www.ijisrt.com 572


Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
connected layer, with a total of approximately 4 million final stage of convolution layers to avoid over-fitting
free parameters. Following each convolutional layer is this model (Conv3-1 and Conv3-2). We used stochastic
batch normalization and a rectified linear activation gradient descent with Nesterov to train the model. The
(ReLU) layer. We include drop out (p = 0.5) in the initial learning rate was set to 0.001, and it was reduced by a
fully connected layer (fc5) and L2 weight factor of 10 if the validation loss did not decrease for four
regularization in the fully connected layer (fc5) and the consecutive epochs.

Fig. 5: The proposed network architecture for simultaneous classification and regression

V. RESULTS AND DISCUSSION step, we trained neural networks to classify manually


annotated knee joint images. After much
The proposed method trained FCNs to automatically experimentation, we arrived at the final architecture
localize and extract knee joints from X-ray images of knee depicted in Figure 5.We compare our network's
OA. The datasets are divided into two parts: classification results to the previous best results for
training/validation (70%) and test (30%). The OAI dataset automatically quantifying knee OA severity, WND-
contains 3,146 images for training and 1,300 images for CHARM, a multipurpose medical image classifier [11].
testing. MOST dataset training and test samples are 2,020 The multi-class classification accuracy and mean-squared
and 900 images, respectively. First, we trained the network error of our network and WND-CHARM are shown in
with OAI dataset training samples before testing it with OAI Table 2. The results show that our network was trained
and MOST datasets separately. Following that, we expanded from the scratch for classifying knee OA images clearly
our training samples by including the MOST training set, outperforms WND-CHARM. Also these results show an
and the test set is a hybrid of the OAI and MOST test sets. These results also show an improvement over our previous
We experimented with the number of convolution stages, reported methods [1], which used off-the-shelf networks like
the number of filters, and the number of layers before VGG nets and the BVLC Reference CaffeNet for classifying
settling on the final architecture. The final network (shown knee OA X-ray images using transfer learning. These gains
in Figure 4) was trained using samples from both the OAI are due to our network's lightweight architecture, which was
and the MOST datasets. trained from scratch with fewer (5.4 million) free parameters
than BVLC CaffeNet, which had 62 million free parameters
A. Classification of Knee OA Images using a CNN for the same amount of training data. The off-the-shelf
We use the same train-test split for localization and networks were trained on a large dataset like ImageNet,
quantification to maintain pipeline consistency and allow which contains millions of images, whereas our dataset
valid comparisons of results obtained using different contains far fewer training samples (10, 000). In the
approaches. To increase the training samples, we include following section, we show further improvements in the
the right-left flip of each knee joint image, which doubles results for quantifying knee OA severity.
the total number of training samples available. As a first

IJISRT22DEC670 www.ijisrt.com 573


Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Network Sensitivity Specificity Accuracy F1-Score


WIND-CHARM method 89.6 79.78 86 90.63
Fractal dimension 92.9 87.88 91.16 93.81
Fine-Tuned BVLC Caffe Net 94.39 88.96 93.85 95.42
Proposed Model 99.651 99.4242 98.755 99.3
Table 1: Comparison of the proposed model with existing system

120

100

80 WIND-CHARM method
60 Fractal dimension
Fine-Tuned BVLC Caffe Net
40
Proposed Model
20

0
Sensitivity Specificity Accuracy F1-Score
Fig. 6: Comparison of proposed work Results with the state of art methods

Fig. 7: (a) Training (Tr) and validation (Val) accuracy (acc), (b) Training and validation loss for joint
classification(Clsf) and regression(Reg) training.

According to the results in Tables 2 and 3, the 0.898. Table 5 compares the precision, recall, F1 score,
network trained jointly for classification and and area under curve (AUC) of the networks trained
regression has a higher multi-class classification for classification and regression. These findings
accuracy of 98.75% and a lower mean-squared error of indicate that the network jointly trained for
0.661 than the previous network trained only for classification and regression learns a better
classification, which has a multi-class classification representation than the previous network trained only
accuracy of 98.75% and a mean-squared error of for classification.

IJISRT22DEC670 www.ijisrt.com 574


Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Fig. 8: Various classification of Osteoarthritis using OAI dataset

From the Figure 10 shows some examples of because the individual categories are not equidistant from
various classifications: knee joints predicted as Normal, each other [3,4]. This could explain why the automatic
Doubtful, Moderate, Mild and Severe. These images show quantification has a low multi-class classification
minimal variations in terms of joint space width and accuracy. Because knee OA features such as joint space
osteophytes formation, making them challenging to narrowing, osteophytes formation, and sclerosis are
distinguish. Even though the KL grades are used in separately graded, using OARSI readings instead of KL
clinical settings to assess knee OA severity, there has grades may provide better results for automatic
been ongoing investigation and criticism of their use quantification.

Fig. 9: Proposed model predicted the given input image as a) Doubtful b) Moderate

IJISRT22DEC670 www.ijisrt.com 575


Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Fig. 10: The obtained Confusion matrix for the proposed model

VI. CONCLUSION osteophytes in the knee doubtful or definite?


Osteoarthritis and cartilage,11(2),149–150(2003).
We proposed new methods for automatically [5.] Jia, Y., Shelhamer, E., Donahue, J., Karayev, S.,
localizing knee joints using a fully convolutional Long, J., Girshick, R., Guadarrama,S., Darrell, T.:
network and quantifying knee OA severity using a Caffe: Convolutional architecture for fast feature
network jointly trained for multi-class classification embedding. In:Proceedings of the ACM International
and regression, both from scratch. In comparison to the Conference on Multimedia. pp. 675–678(2014)
previous methods, the FCN-based method is highly [6.] Karayev, S., Trentacoste, M., Han, H., Agarwala, A.,
accurate. We demonstrated that the classification Darrell, T., Hertzmann,
results obtained with automatically localized knee A.,Winnemoeller,H.:Recognizingimagestyle.arXivpre
joints are comparable to the results obtained with printarXiv:1311.3715(2013)
manually segmented knee joints. In comparison to the [7.] Kingma, D., Ba, J.: Adam: A method for stochastic
previous method, the jointly trained network for optimization. arXiv preprintarXiv:1412.6980(2014)
classification and regression improves in multi-class [8.] Liu, S., Yang, J., Huang, C., Yang, M.H.: Multi-
classification accuracy, precision, recall, and F1 score. objective convolutional learningfor face labeling. In:
The confusion matrix and other metrics show that Proceedings of the IEEE Conference on Computer
classifying Knee OA images conditioned on KL grade 1 Vision andPatternRecognition.pp.3451–3459(2015)
is difficult due to small variations, especially in the [9.] Long, J., Shelhamer, E., Darrell, T.: Fully
consecutive grades from grade 0 to grade 2. Future work convolutional networks for
will focus on training an end-to-end network to quantify semanticsegmentation.In:ProceedingsoftheIEEEConf
knee OA severity by integrating the FCN for localization erenceonComputerVisionandPatternRecognition.pp.3
and the CNN for classification. It will be interesting to 431–3440(2015)
compare the human-level accuracy involved in assessing [10.] Oka, H., Muraki, S., Akune, T., Mabuchi, A., Suzuki,
knee OA severity to automatic quantification methods. T., Yoshida, H., Yamamoto,S., Nakamura, K.,
This could provide insights into how to improve fine- Yoshimura, N., Kawaguchi, H.: Fully automatic
grained classification even further. quantificationof knee osteoarthritis severity on plain
radiographs. Osteoarthritis and
REFERENCES Cartilage16(11),1300–1306(2008)
[1.] Antony, J., McGuinness, K., Connor, N.E., Moran, [11.] Orlov, N., Shamir, L., Macura, T., Johnston, J.,
K.: Quantifying radiographic knee osteoarthritis ever Eckley, D.M., Goldberg, I.G.: WND-CHARM:
it using deep convolutional neural networks .In: Multi-purpose image classification using compound
Proceedings of the 23rd International Conference on image
Pattern Recognition. IEEE(2016),InPress. transforms.Patternrecognitionletters29(11),1684–
[2.] Chatfield, K., Simonyan, K., Vedaldi, A., Zisserman, 1693(2008)
A.: Return of the devil in the details: Delving deep [12.] Park,H.J., Kim, S.S., Lee, S.Y., Park, N.H., Park,
into convolutional nets. In: Proceedings of British J.Y., Choi, Y.J., Jeon, H.J.:A practical MRI
Machine Vision Conference(2014). grading system for osteoarthritis of the knee:
[3.] Emrani, P.S., Katz, J.N., Kessler, C.L., Reichmann, association withKellgren–Lawrence radiographic
W.M., Wright, E.A., McAlindon,T.E., Losina, E.: scores. European journal of radiology 82(1), 112–
Joint space narrowing and Kellgren–Lawrence 117(2013)
progression in knee osteoarthritis: an analytic [13.] Russakovsky,O.,Deng, J., Su, H., Krause, J.,
literature synthesis. Osteoarthritis and Cartilage Satheesh, S., Ma, S., Huang,Z., Karpathy, A.,
16(8),873–882(2008). Khosla, A., Bernstein, M., et al.: ImageNet large
[4.] Hart,D.,Spector,T.: Kellgren & Lawrence grade1 scale visualrecognition challenge. International

IJISRT22DEC670 www.ijisrt.com 576


Volume 7, Issue 12, December – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Journal of Computer Vision 115(3), 211–252(2015)
[14.] Shamir,L.,Ling,S.M., Scott, W., Hochberg, M.,
Ferrucci, L., Goldberg, I.G.:Early detection of
radiographic knee osteoarthritis using computer-aided
analysis.OsteoarthritisandCartilage17(10),1307–
1312(2009)
[15.] Shamir, L., Ling, S.M., Scott Jr, W.W., Bos, A.,
Orlov, N., Macura, T.J.,
Eckley,D.M.,Ferrucci,L.,Goldberg,I.G.:KneeX-
rayimageanalysismethodforautomateddetectionofoste
oarthritis.IEEETransactionsonBiomedicalEngineering
56(2),407–415(2009)
[16.] Shamir, L., Orlov, N., Eckley, D.M., Macura, T.,
Johnston, J., Goldberg, I.: Wnd-charm:Multi-
purposeimageclassifier.AstrophysicsSourceCodeLibr
ary(2013)
[17.] Shamir, L., Orlov, N., Eckley, D.M., Macura, T.,
Johnston, J., Goldberg, I.G.:Wndchrm–an open
source utility for biological image analysis. Source
code forbiologyandmedicine3(1),13(2008)
[18.] Simonyan, K., Zisserman, A.: Very deep
convolutional networks for large-
scaleimagerecognition.arXivpreprintarXiv:1409.1556
(2014)
[19.] Yang, S.: Feature Engineering in Fine-Grained Image
Classification. PhD
Thesis,UniversityofWashington(2013)
[20.] Yoo, T.K., Kim, D.W., Choi, S.B., Park, J.S.: Simple
scoring system and
artificialneuralnetworkforkneeosteoarthritisriskpredic
tion:Across-
sectionalstudy.PloSone11(2),e0148724(2016

IJISRT22DEC670 www.ijisrt.com 577

You might also like