2019 International Conference on Computational Intelligence and Knowledge Economy (ICCIKE)

December 11–12, 2019, Amity University Dubai, UAE

Detection of Rheumatoid Arthritis Using Machine

Learning
Utkarsh Vikram Singh Eva Gupta, Tanupriya Choudhury,
Dept. of Informatics, School of CS, Dept. of Informatics, School of CS, Dept. of Informatics, School of CS,
University of Petroleum and Energy University of Petroleum and Energy University of Petroleum and Energy
Studies, Dehradun Studies, Dehradun, Studies, Dehradun,
[email protected], [email protected] [email protected]

Abstract: Artificial neural network is inspired by the way the bodies which in turn encourages us to fight the infection, so
biological neural system works i.e. how the brain processes when somebody is detected with rheumatoid arthritis (RA),
information. AI and ML is being trialed for a range of research their immune system by unintentionally transmits the
and healthcare uses, including detection of various kind of antibodies to the covering of the joints where they attack the
disease, management of chronic(persistent) conditions, delivery tissue of the joint, causing the layers of joints to become sore
and discovery of health services, and drug respectively. and also releasing chemicals which destroys bones,
Rheumatic diseases are more common than any other type of cartilages, ligaments, tendons. If early treatment is not
diseases, rheumatism or we can say the musculoskeletal provided it can cause the joint to lose its original shape and
system’s pain that affects the life’s day to day activities. It is of alignment. Early findings and cure for rheumatoid arthritis
great importance to diagnose patients that are more vulnerable (RA) can increase the healing possibility. So many factors are
to rheumatic diseases in terms of life quality. It targets all ages there so as to detect rheumatoid arthritis in which the values
but it is more common in women. This disease has many of the Anti CCP, Rheumatoid Factor (RF), ESR and SJC are
symptoms similar to other diseases. Therefore, it is very hard to most prominent in order to detect rheumatoid arthritis (RA).
detect. Also, the diagnostic tools are complex and uneconomical.
In this paper, the clustering techniques are used, a method of
In this research paper, machine learning algorithms are
representation of unsupervised learning, to expect rheumatic
implemented to predict rheumatic arthritis (RA) by the help of
disease patients as early as possible.
the four factors for the study of rheumatic diseases. In the
upcoming future, using the artificial intelligence (AI), it aids to
enhance the prediction of rheumatic diseases. SYMPTOMS OF RHEUMATIC DISEASES
The pain in the joints is the foremost symptom of RA, leading
Keywords: Machine Learning (ML), Musculoskeletal to bulge and stiffness. It may cause other general symptoms
System, Rheumatism, Rheumatic Arthritis (RA) and inflammation in different parts of the human body which
appears in some weeks but some other may progress quickly
I. INTRODUCTION in few days only. People may also experience flares when the
condition begins to deteriorate and post symptoms become
In northern part of India, in the past the social and economic more severe.
environment seems less better in some area, the happening
rate of rheumatic diseases goes on to increase due to maturity. It also causes problems in joints, although the smaller joints
When people enter the aging generation, the chances of the in the palm, fingers and feet especially toes are often prone
number of diseases increases which in turn depletes the QOL. to be affected first. More than two third of the patient
The treatment of RA disease leads to the rise of cost socially experience fatigue, Ambiguous Muscle and Anorexia. Other
and economically. traits which people suffering from RA face are sweating,
The idea of healthy underlines the quality of life (QOL) or weight loss, laziness, lack of energy and high temperature
the satisfaction of the person, not just the disease. The quality
that is fever. These signs are tough to diagnose the rheumatic
of life (QOL) is redefined by the World Health Organization
diseases. The doctor's diagnosis this disease with the help of
(WHO) as an acknowledgement level in the culture and way
of life in which the people live in. Particularly, rheumatoid laboratory findings, physical examinations, and imaging. The
and osteoarthritis arthritis brought by maturing takes the diagnosis of RA is done by many steps and procedures. It is
second spot in the analysis of such chronic diseases. Though significant to foretell the RA diseases in advance in order to
the accurate and genuine cause of rheumatoid arthritis (RA) boost the elderly’s QOL as they are aging.
is not explained at this point yet in a complicated manner
most likely environmental and hereditary factors interact. It
is a prolonged sickness that affects more than one areas in the
joint [1]. The immune system produces the antibodies which
attacks the bacteria and dangerous viruses entering into the

978-1-7281-3778-0/19/$31.00 ©2019 IEEE

25

Authorized licensed use limited to: University of Canberra. Downloaded on June 06,2020 at 15:36:22 UTC from IEEE Xplore. Restrictions apply.
 2019 International Conference on Computational Intelligence and Knowledge Economy (ICCIKE)
December 11–12, 2019, Amity University Dubai, UAE

TABLE I. THE CRITERION FOR THE RHEUMATOID ARTHRITIS (RA) Hierarchical “nested” clusters produces many other nested
CLASSIFICATION
clusters in the period. The relevance of clustering is varied
CRITERION DESCRIPTION and also helpful in many other fields. In cluster analysis, we
segment the four factors into the four variables[2, 3]. The K-
means clustering is also a representative way of segmenting
A. Morning It occurs inside and surrounding the joints. cluster analysis. The K-Means clustering is a widely used in
severity It goes for at least 1 hour before maximum unsupervised learning algorithm. A centroid is preferred for
improvement happens. the K-Means clustering algorithm, then the classification is
B. Arthritis of 3 The 14 possible areas are left or right PIP,
done by making the application of the space between the
or more joint wrist, ankle elbow, knee, MTP and MCP centroid and data following it [4]. The subsequent step is to
areas
joints. recur the process and then arranging of the centroids based
on the more centrally located centroid, and thus concludes if
there is no further possibility of cluster classification. The By
decreasing the distance say Euclidean from the main center
C. Hand joints - Minimum one part of area is swollen in a point and then doing the early erroneous merging algorithm
arthritis MCP or PIP joints.
frequently, the K-means clustering can be regenerated [5].
D. Symmetric Continuous participation of the same areas Therefore, a meaningful structure of the data gathered can be
arthritis of the joint (as defined above in B) on both calculated without any prior knowledge of the data.
the sides (left and right) of the human
body (bilateral involvements of the MCPs, Algorithm of “K-means” clustering:
PIPs, or MTPs is admissible without any 1: Initiate
proper symmetry) 2: Select the count of clusters ‘K’.
E. Rheumatoid Hypodermic nodules, extensor region or 3: Following step is to select randomly the K points that are
nodules over bone spur and articular part of juxta. centroids not necessarily from the dataset.
4: Now, form a cluster of each of the data points to the
F. Serum Exhibition of varied quantity of serum adjoining centroids so as to form the clusters possible.
“Rheumatoid rheumatoid factor by any method 5: Calculate and place again the new centroid of each of the
Factor (RF)” regarding which the outcome is mostly cluster so as to find most accurate results.
certain and has been positive in less than 6: Rearrange and assign each of the data point to the nearest
5% of ordinary control motif (OCM).
centroid and if any reassignment among took place go to
G. Radiographic Radiographic Variation (RV) point number 5.
Variation quintessential of (RA) on wrist and poster
anterior part of hand, which should include
7: Terminate
unambiguous bony decalcification restrain
in to the participating joints, Algorithm of Hierarchical clustering:
“osteoarthritis” variation alone cannot 1: Initial step is to build each data point into a singular point
pass. cluster -> forming “N” clusters.
2: Take the two nearest cluster and combine them into one
single cluster -> forming “N-1”number of clusters.
3: Chose the two closest clusters and make them into one
CLUSTERING cluster -> that forms N-2 number of clusters.
Clustering is a machine leaning technique which comes 4: Step number 3 is repeated until there is one cluster only.
5: Finish
under unsupervised learning. It involves grouping of data
points. An unsupervised learning method is a method in
Algorithm of DBSCAN clustering:
which there is no teacher to train as compared to supervised
learning, the structures are formed by visualizing the dataset 1: DBSCAN(dataset, eps, MinPts){
and finding appropriate groups of items. It’s a process to find 2: C = 1.
a meaningful groups and in turn form the structure. 3: In dataset, for each and every unvisited point “p” {
Clustering is the work done by dividing the population into 4: mark point “p” as visited.
a large number of groups in such a way that data points in the 5: Neighbors N = finds the adjacent points of “p”
similar groups are more alike the other points in the same 6: if |N|>=MinPts:-
group and unlike the data points in other such groups. It is a 7: N = N U N'
collection of groups on the basis of similarity and 8: if point “p” does not belong to any cluster:
9: add point “p” to cluster “C”}
dissimilarity between them which forms clusters.
There are three significant clustering methods. The K-means
clustering algorithm the specified data is clustered into
clusters k, decreasing the space separation which is variant
with each and every cluster. In the Hierarchical clustering,
the data is clustered as N constituent factors into M clusters.

26
Authorized licensed use limited to: University of Canberra. Downloaded on June 06,2020 at 15:36:22 UTC from IEEE Xplore. Restrictions apply.
 2019 International Conference on Computational Intelligence and Knowledge Economy (ICCIKE)
December 11–12, 2019, Amity University Dubai, UAE

TABLE II. EXAMPLE OF DATA SET FOR PREDICTIVE To foretell the “rheumatic diseases”, anonymous data of 60
patients were clustered randomly into four varieties of
attributes. We choose the K-means clustering algorithm in
order to prove the reliability in the outcomes of upcoming
studies with that of the patients suffering from RA (Fig. 1).
Although the data gathered is comparatively small, 4 of the
60 parameters are randomly clustered into 4 groups. Four
clusters were made. The comparative analysis results using
four factors that’s why 4 clusters were selected. Fig. 2 shows
the K-means model visualization which is python program
based.
Similar results can be predicted for factors rheumatoid factor,
antiCCP, EBS as well.
We have compared each factor with age and analyzed which
factor is more dominant in what age group as shown in Fig.
2 and 3.
IV. RESULTS AND DISCUSSION

III. METHODOLOGY
In the study, four of the factors among all were diagnosed.
Patients with RA were analyzed for the following:-
rheumatoid factor (RF) > 7, Swollen Joint Count (SJC) > 4,
anti-cyclic citrullinated peptide (CCP) > 18, and ESR> 25. F
1 Table II shows data of the patient which included the above
specified factors.
Fig. 2. Execution of K-means Clustering (Part 2)

Fig. 1. Execution of K-means (Part 1)

27
Authorized licensed use limited to: University of Canberra. Downloaded on June 06,2020 at 15:36:22 UTC from IEEE Xplore. Restrictions apply.
 2019 International Conference on Computational Intelligence and Knowledge Economy (ICCIKE)
December 11–12, 2019, Amity University Dubai, UAE

Fig. 3. Execution of Hierarchical Clustering

Real life data (rheumatoid arthritis) possibly will have

specific irregularities:–
i) The Clusters might be of an arbitrary shape. Fig. 5: Sensitivity towards outlier
ii) Data may contain noise. For dealing with such kind of
data DBSCAN comes into picture.
3. K-means and hierarchical clustering algorithms
Time complexity of DBSCAN and k-means is similar (O (agglomerative) need one to specify the count of
(n2)) whereas the time taken by hierarchical algorithm is
clusters to be made whereas DBSCAN algorithm
more (O (n3)).
doesn’t require that. DBSCAN algorithm identifies
DBSCAN < K-means < Hierarchical the dense region made by grouping together the data
“K-means” algorithm face major problems in identifying points which are near to each other on the basis of the
the clusters with random shapes. space measured between them.

TABLE III. SHOWS A COMPARATIVE STUDY HIGHLIGHTED THE ABOVE

Disadvantage of K-means Algorithm and Hierarchical KEY POINTS

Algorithm [6] Parameter K-Means Hierarchical DBSCAN

/Algorithm Clustering Clustering Clustering
Algorithm Algorithm Algorithm
(AGGLOMERATIVE) clustering algorithms-
1. K-Means and Hierarchical algorithms form the Time O(n2) O(n³) O(n2)
spherical shaped clusters only. Both of the algorithms
Estimated time 0.22099995613 0.352999925 0.211999893
fails to succeed when data taken is not of spherical of execution (for 1 613 188
shape (i.e. similar variance in all the directions). a dataset of 100
entries)

Space O((m+k)n) O(n²) O(n)

V. CONCLUSION
DBSCAN algorithm has quite a lot of advantages over the
hierarchical clustering and K-means clustering in areas
where there is a need of handling and representing data.
Partitioning methods (K-means) and hierarchical
clustering (Agglomerative) represent the data in the form
of convex or spherical-shaped clusters. We can also
conclude that they are more suitable for well-separated
and small i.e. compact clusters as they are drastically
affected by the outliers and noise in the data set used
above.

Fig. 4. DBSCAN and K-means comparison

2. K-Means and hierarchical Algorithms are sensitive

towards outliers. Outliers can skew the clusters in K-
Means in very large extent and the algorithms won’t
work as expected to perform.

28
Authorized licensed use limited to: University of Canberra. Downloaded on June 06,2020 at 15:36:22 UTC from IEEE Xplore. Restrictions apply.
 2019 International Conference on Computational Intelligence and Knowledge Economy (ICCIKE)
December 11–12, 2019, Amity University Dubai, UAE

us in the more accurate and reliable diagnosis of

Estimated time of rheumatoid arthritis.

0.4
execution REFERENCES
[1] Henk Visser, Saskia le Cessie, et al “How to Diagnose Rheumatoid
Arthritis Early”, ARTHRITIS & RHEUMATISM, Vol. 46, No. 2
(2002), DOI 10.1002/art.10117, 357–365.
0.2
[2] Yong Gyu Jung, et al “Clustering performance comparison using
Y axis: seconds

K-means and expectation maximization algorithms”,

0 Biotechnology & Biotechnological Equipment, Vol. 28, No. S1
(2014), 45-48, http://dx.doi.org/10.1080/13102818.2014.949045.
K means Hierachial DBSCAN
Series1 [3] Min-Soo Kang, et al, “A Study on the Search of Optimal
X axis:Clustering Algorithms Aquaculture farm condition based on Machine Learning”, The
Journal of The Institute of Internet, Broadcasting and
Communication (IIBC) Vol. 17, No. 2 (2017), 135140,.
Fig. 6. Bar graph representation of execution time
[4] Jae-Gyun Park, et al, “Dropout Genetic Algorithm Analysis for
The study of rheumatoid arthritis is a kind of study which Deep Learning Generalization Error Minimization”, International
has a scope of more observation. Clustering helps in Journal of Advanced Culture Technology Vol.5 No.2 (2017), 74-
81, https://doi.org/10.17703/IJACT.2017.5.2.74.
getting the clusters of the significant groups and factor
domination age wise. With the help of clustering [5] Beom-Joo Park, et al, “A Study on Efficient Memory Management
algorithms we can get the comparison of the 3 algorithms Using Machine Learning Algorithm”, International Journal of
on the basis of their execution time. Now we can further Advanced Smart Convergence Vol.6 No.1 (2017), 39-43,
https://doi.org/10.7236/IJASC.2017.6.1.39.
it with the help of more precise study of the algorithms of
[6] Debomit Dey, “DBSCAN Clustering in ML”, geeksforgeeks.com
machine learning and improvising them which can help [online], Available: https://www.geeksforgeeks.org/dbscan-
clustering-in-ml-density-based-clustering/.

29
Authorized licensed use limited to: University of Canberra. Downloaded on June 06,2020 at 15:36:22 UTC from IEEE Xplore. Restrictions apply.