Content uploaded by Ahmet Çifci
Author content
All content in this area was uploaded by Ahmet Çifci on Jul 19, 2019
Content may be subject to copyright.
184
A Study on Method Prediction for a Better Directed Treatment of Warts
Ahmet ÇİFCİ1, Mehmet ŞİMŞİR2
1Deparment of Electrical-Electronics Engineering, Burdur Mehmet Akif Ersoy University,
TURKEY
2Department of Mechatronics Engineering, Karabük University, TURKEY
Abstract: Various kinds of medical treatment methods may be used to cure common types of
diseases. Experience based predictions are done to choose a treatment method among the
choices of cures to get better results for the patient. This condition sometimes may continue
with trying another treatment method unless a satisfactory result is reached and changing the
treatment method of the cure process is not a desired course for time and health. This study
presents a confident way to choose the treatment method for wart disease by using feedforward
neural network. The study uses two types of datasets, one for cryotherapy and other for
immunotherapy treatment methods. It was observed from the experimental results that,
feedforward neural network achieved 94.4% success and 85.6% success for cryotherapy and
immunotherapy datasets, respectively. The results are remarkable for both doctors and patients.
Keywords: Cryotherapy, Immunotherapy, Feedforward Neural Network, Wart.
INTRODUCTION
Warts are benign skin growth caused by the virus called Human papillomavirus (HPV), which
infect the top layer of the skin [1]. Most warts are harmless, but they are highly contagious.
Traumas events such as cuts or damages to the skin facilitate infection. They are spread so
quickly, especially in the summer. Direct contact with a wart or contacts with another person
by sharing towels, razors, or other personal items may cause the virus to spread [2]. Although
they are most common on the knuckles, fingers, hands, elbows, knees, they may occur in the
whole body. Children, adolescents, people who bite their nails, and people with a weak immune
system have a higher risk of developing warts [3]. Warts are usually flesh-coloured, hard and
rough. On the other hand, there are dark (brown, grey-black), flat and soft wart types.
185
Types of wart include common warts (verruca vulgaris), plantar warts (verruca
plantaris), flat warts (verruca plana), and genital warts (condylomata acuminata) [4]. Common
warts are most commonly seen in the hands, between the fingers, legs, and around the nails [5].
Plantar warts are embedded in the skin and can be painful. Plantar warts can be confused with
calluses [6]. Even though flat warts can occur on any part of the body, they are often found on
the face and hands. They are small, smooth and flesh-coloured. They can occur in large numbers
[4]. Genital warts are soft, flesh‐coloured papules on the genitalia and breech. It is more
common in people who have sexual activity without a condom [4].
There exist several methods available for the treatment of warts. Salicylic acid
treatment, electrosurgery, freezing (cryotherapy), immunotherapy, and laser treatment are the
main methods to get rid of warts [7]. The most common methods for the treatment of warts are
cryotherapy and immunotherapy. Cryotherapy is performed by applying liquid nitrogen gas to
the desired skin lesion using special devices. Liquid nitrogen gas at a temperature of -196 degree
Celsius will freeze the tissue. This freezing process is a short duration of time (10 - 60 seconds);
at the end of this time, the tissue will return to its normal temperature. In this short-term and
immediate freezing-melting process, the cells in the targeted tissue will be destroyed and will
die. These dead and abnormal cells, which no longer function, will be removed from the tissue
during the healing process and replaced by fresh tissues [8,9]. Immunotherapy is a new type of
treatment that aims to prevent or eliminate the growth of cancer cells that abnormally proliferate
(such as HPV lesions). Immunotherapy uses the patient's own immune system to fight warts.
The antigen is injected into the body to activate the immune system [10,11]. Cryotherapy is an
uncomplicated method for the patients with respect to immunotherapy.
Numerous investigations have been reported in the literature on the treatment of warts.
In [12], Putra et al. used AdaBoost and Random Forest as a strong learner or a weak learner for
selection of wart treatment method. The results showed that the accuracy rate of cryotherapy
was 96.6% and accuracy rate of immunotherapy was 91.1%. Tanyıldızı et al. [13] judged the
performances of classification algorithms in cryotherapy and immunotherapy datasets. The best
success rate obtained using the K-star algorithm is found as 96.66% for cryotherapy and using
the Random Forest algorithm is found as 85.55% for immunotherapy. Fuzzy Rule, Naive Bayes,
and Random Forest based algorithms have been carried out in cryotherapy and immunotherapy
treatment for comparing the effectiveness of these algorithms by Akyol et al. in [14]. They
186
concluded that the random forest algorithm outperforms other classification algorithms in both
accuracy and sensitivity within cryotherapy and immunotherapy datasets. Nugroho et al.
proposed C4.5 algorithm combined with Random Forest Feature Weighting for wart treatment
selection method [15]. The results showed that the proposed method can improve the
performance of prediction. Ali et al. applied some algorithms to show that which treatment is
more effective from cryotherapy and immunotherapy [16]. They concluded that cryotherapy
treatment is better than immunotherapy. See also [17], where the author used the decision tree-
based method to specify the rules of predicting the performance of wart treatment methods. The
results obtained the level of accuracy of 94.4% on cryotherapy and 90% on immunotherapy.
In this study, feed-forward neural network was used to decide and select the cure for
wart treatment. Patient data were used as input and success of cure type was used as output
data. A neural network for immunotherapy and another one for cryotherapy was built to decide
which methodology is appropriate for the patient. As it was mentioned before, cryotherapy is
an uncomplicated method for the patient with respect to immunotherapy. When a patient data
is applied to neural network for cryotherapy and gives positive outputs for applicability.
Treatment method starts with cryotherapy cure, but if cryotherapy neural network gives
ineffective result as an output, applying data to immunotherapy neural network is preferred and
this situation supports a logical way to select the uncomplicated method at first without waste
of time. Because immunotherapy method has a higher result of success percentage according
to the datasets used in this study. But this method may also be abrasive for the patient.
The paper is structured in the following manner. Section 2 presents the immunotherapy
and the cryotherapy datasets. Method and experimental results are given in section 3. Finally,
conclusions are duly drawn in section 4.
DATASET
The immunotherapy and the cryotherapy datasets used in this study are gathered from the
University of California, Irvine (UCI) Machine Learning Repository [18,19]. The datasets are
collected along two years from the dermatology clinic of Ghaem Hospital in Mashhad, Iran
[20,21]. The immunotherapy with candida antigen and the cryotherapy with liquid nitrogen
were applied to 180 patients with plantar and common warts. Each data set contains 90 patients.
Patients were randomly selected. The datasets do not have any missing value.
187
The immunotherapy dataset consists of eight features; gender, age, time elapsed before
treatment, the number of warts, types of wart, surface area of warts, induration diameter of
initial test, and response to treatment. The details of the immunotherapy dataset are presented
in Table 1.
Table 1. Features of Immunotherapy Dataset
Feature No.
Feature Name
Values
1
Gender
41 Male
49 Female
2
Age (year)
15-56
3
Time elapsed before treatment (month)
0-12
4
The number of warts
1-19
5
Types of wart (count)
47 Common
22 Plantar
21 Both
6
Surface area of warts (mm2)
6-900
7
Induration diameter of initial test (mm)
5-70
8
Response to treatment
Yes or No
The cryotherapy dataset has seven features; gender, age, time elapsed before
treatment, the number of warts, types of wart, surface area of warts, and response to
treatment. The details of the cryotherapy dataset can be seen in Table 2.
188
Table 2. Features of Cryotherapy Dataset
Feature No.
Feature Name
Values
1
Gender
47 Male
43 Female
2
Age (year)
15-67
3
Time elapsed before treatment (month)
0-12
4
The number of warts
1-12
5
Types of wart (count)
54 Common
9 Plantar
27 Both
6
Surface area of warts (mm2)
4-750
7
Response to treatment
Yes or No
METHOD AND EXPERIMENTAL RESULTS
With the developments on intelligent systems, human expertise can be simulated by artificial
neural networks to gain time and to get more accurate decisions. As a general approach, our
study is based on previously taken and reliable reference data. After a general overview on
types of cures for wart treatment, it is necessary to classify the patient data. “Gender”, “age”,
“time elapsed before treatment”, “the number of warts”, “types of wart” and “surface area of
warts” features for the patients were used as the classification titles for the neural network data
inputs. And output data was also used as success of cure type as “0” and “1”.
A feedforward neural network was chosen for this study. Feedforward neural networks
has input, output and at least one hidden layers. The neural network was constructed with one
hidden layer. If too many hidden layers and neurons are used, overfitting may occur, that is
although the network can be trained to work very well for the training data, it performs poorly
for test data. It is essential to optimize the success percentage of the neural network with
effectual numbers of neurons for the hidden layer to achieve the best performance level
[22,23].
189
The constructed neural networks have 6 input variables, 6 neurons for the hidden layer
and also 2 for output for both cryotherapy and immunotherapy as seen in Fig. 1.
6 Input Variables 6 neurons 2 neurons
Input Layer Hidden Layer Output Layer
Inputs
Outputs
Fig. 1. General Structure Of Constructed Feed Forward Backpropagation Neural Networks
For Both Cryotherapy And Immunotherapy.
To determine the best condition for a better success percentage of neural networks,
several numbers of neuron numbers were experimented by MATLAB and current configuration
shown in Fig. 1 was obtained because of better success rate according to the tested other
choices. The training performance curves of the designed neural networks are shown in Fig. 2
for cryotherapy and Fig. 3 for immunotherapy. Best validation level of the neural networks for
training process is indicated by means of training, validation and test curves. The training
process of the neural network was reached to goal at 7 epochs in terms of minimum gradient
for cryotherapy neural network.
190
Fig. 2. Training Performance Curves of Neural Network Designed For Cryotherapy
by MATLAB.
Fig. 3. Training Performance Curves of Neural Network Designed For Immunotherapy
By MATLAB.
The training state throughputs of the neural network are shown in Fig. 4 for cryotherapy
and Fig. 5 for immunotherapy, which indicate weight changes and validation checks for the
neural networks and gradients till the determined goals were reached.
191
Fig. 4. Training State Data of Neural Network For Cryotherapy by MATLAB.
Fig. 5. Training State Data of Neural Network For Immunotherapy by MATLAB.
After acceptable satisfactory levels were reached according to the number of data used,
it was necessary to test the success rate of the designed and trained neural networks by the help
of MATLAB Simulink model as seen in Fig. 6. The system in Fig. 6 was designed for testing
the constructed neural networks using the pre-paired patient data to achieve a performance test.
192
Fig. 6. MATLAB Simulink Model For Neural Network Performance Test.
After performance test for cryotherapy neural network, a success percentage of 94.44%
was achieved. And the neural network test results for immunotherapy was 85.6% after the same
performance test which was realized using the model seen in Fig. 6. Both cryotherapy and
immunotherapy network trainings were practiced using 90 sets of patient data but there is a
success percentage difference between them. The cause of this difference is the success level
of immunotherapy method. Immunotherapy is more successful with respect to cryotherapy
according to the datasets used in this study, it is also a more onerous cure for wart treatment.
And this rate of success for cure treatment decreases the variety of outputs as a result reduction.
As a result of this situation, the success percentage of the neural network was directly affected.
As it was mentioned before, to find a suitable cure for wart treatment is important. And
it is also a noteworthy point that; trying the easier and more effortless cure for the patient is a
big advantage. Although the onerous method has a lower level of neural network success
percentage. There is a confident way to pre-test the cure type with the neural network to prevent
loss of time. This situation will also increase the success percentage and usage numbers of the
more effortless cure types for treatments by pre-testing them firstly instead of directly using
onerous cures of treatments.
CONCLUSION
In this study, the most common wart treatment methods, cryotherapy and immunotherapy, were
analysed for wart treatment prediction by applying a feedforward neural network. The
experimental results show that the successes of feedforward neural network were 94.4% and
193
85.6% for cryotherapy and immunotherapy methods respectively. Although acceptable levels
of success percentages were obtained by the designed neural networks, it is also possible to
achieve better levels of success percentages for the neural networks using more patient data
while training the designed neural networks.
This study takes an inspiring role in the cure selecting for wart treatment by obtaining
positive results for a better directed preference. Using artificial intelligence for cure selections
of treatments will cause obtaining faster and well directed treatment processes.
REFERENCES
[1] K. E. Rowson, and B. W. Mahy, “Human papova (wart) virus,” Bacteriological Reviews,
vol. 31, no. 2, pp. 110–131, 1967.
[2] D. C. Wiley, and A. C. Cory, Encyclopedia of School Health, Thousand Oaks, CA: SAGE
Publications, 2013.
[3] S. K. Loo, and W. Y. Tang, “Warts (non-genital),” BMJ Clinical Evidence, 1710, 2009.
[4] L. Allison, and M. D. Holm, “Warts,” in Pediatric Clinical Advisor, vol. II, L. C. Garfunkel,
J. Kaczorowski, and C. Christy, Eds. St. Louis: Mosby Elsevier, 2007.
[5] W. Bonnez, “Papillomaviruses,” in Mandell, Douglas, and Bennett’s Principles and Practice
of Infectious Diseases, vol. VIII; J. E. Bennett, R. Dolin, and M. J. Blaser, Eds. Philadelphia: Saunders,
an Imprint of Elsevier Inc., 2015.
[6] S. M. Simons, And R. Kennedy, “Foot Injuries,” in Clinical Sports Medicine Medical
Management And Rehabilitation, W. R. Frontera, S. A. Herring, L. J. Micheli, And J. K. Silver, Eds.
Philadelphia: Saunders Elsevier, 2007.
[7] N. H., Boroujeni, And F. Handjani, “Cryotherapy Versus Co2 Laser in The Treatment of
Plantar Warts: A Randomized Controlled Trial,” Dermatology Practical & Conceptual, Vol. 8, No. 3,
Pp. 168-173, 2018.
[8] S. Cockayne, C. Hewitt, K. Hicks, S. Jayakody, A.R. Kang’ombe, E. Stamuli, Et Al.
“Cryotherapy Versus Salicylic Acid For The Treatment Of Plantar Warts (Verrucae): A Randomised
Controlled Trial,” Bmj, 342, P. D3271, 2011.
194
[9] M. Barrett, N. Harter, And A. Wysong, “Common Pediatric Dermatologic Surgery
Procedures,” in Pediatric Dermatologic Surgery, K. Nouri, L. Benjamin, J. Alshaiji, And J. Izakovic,
Eds. Hoboken: Wiley & Blackwell, 2019, Pp. 73–92.
[10] Y. C. T. Shih, L. S. Elting, A. L. Pavluck, A. Stewart, And M. T. Halpern, “Immunotherapy
İn The İnitial Treatment Of Newly Diagnosed Cancer Patients: Utilization Trend And Cost Projections
For Nonhodgkin's Lymphoma, Metastatic Breast Cancer, And Metastatic Colorectal Cancer,” Cancer
Invest, Vol. 28, No. 1, Pp. 46–53, 2010.
[11] S. Garg, And S. Baveja, “Intralesional İmmunotherapy For Difficult to Treat Warts With
Mycobacterium W Vaccine,” Journal Of Cutaneous And Aesthetic Surgery, Vol. 7, No. 4, Pp. 203–208,
2014.
[12] M. A. Putra, N. A. Setiawan, And S. Wibirama, “Wart Treatment Method Selection Using
Adaboost With Random Forests As A Weak Learner,” Communications in Science And Technology,
Vol. 3, No. 2, Pp. 52–56, 2018.
[13] E. Tanyıldızı, M. Karabatak, G. Yıldırım, And Z. Özpolat, “Siğil Tedavisinde Sınıflandırma
Algoritmalarının Performans Analizi,” Fırat Üniversitesi Mühendislik Bilimleri Dergisi, Vol. 30, No. 2,
Pp. 249– 256, 2018.
[14] K. Akyol, A. Karacı, And Y. A. Gültepe, “Study On Prediction Success of Machine
Learning Algorithms For Wart Treatment,” International Conference on Advanced Technologies,
Computer Engineering And Science, May 11-13, 2018 Safranbolu, Turkey, Pp. 186–188.
[15] H. W. Nugroho, T. B. Adji, And N. A. Setiawan, “Random Forest Weighting Based Feature
Selection For C4.5 Algorithm on Wart Treatment Selection Method,” International Journal On
Advanced Science, Engineering And Information Technology, Vol. 8, No. 5, Pp. 1858–1863, 2018.
[16] Y. Ali, M. Shahzad, And A. Akbar, “Comparison Of Cryotherapy and Immunotherapy in
Warts Treatment,” International Journal Of Multidisciplinary Sciences And Engineering, Vol. 9, No. 7,
Pp. 14–18, 2018.
[17] S. B. Akben, “Predicting The Success Of Wart Treatment Methods Using Decision Tree
Based Fuzzy İnformative Images,” Biocybern. Biomed. Eng., Vol. 38, Pp. 819–827, 2018.
[18] Https://Archive.İcs.Uci.Edu/Ml/Datasets/Immunotherapy+Dataset
[19] Https://Archive.İcs.Uci.Edu/Ml/Datasets/Cryotherapy+Dataset+
195
[20] F. Khozeimeh, R. Alizadehsani, M. Roshanzamir, A. Khosravi, P. Layegh, And S.
Nahavandi. “An Expert System For Selecting Wart Treatment Method,” Computers in Biology And
Medicine, Vol. 81, Pp. 167–75, 2017.
[21] F. Khozeimeh, F. Jabbari Azad, Y. Mahboubi Oskouei, M. Jafari, S. Tehranian, R.
Alizadehsani, and P. Layegh, “Intralesional İmmunotherapy Compared To Cryotherapy in The
Treatment of Warts,” International Journal Of Dermatology, Vol. 56, Pp. 474–478, 2017.
[22] L. Wang, H. C. Quek, Hou, K. H. Tee, N. Zhou, And C. Wan, “Optimal Size Of A
Feedforward Neural Network: How Much Does it Matter?,” Joint International Conference on
Autonomic And Autonomous Systems And International Conference On Networking And Services
Icas/Icns, October 23-28, 2005 Papeete, Tahiti, French Polynesia, P. 69.
[23] M. Şimşir, R. Bayır, And Y. Uyaroğlu, “Real-Time Monitoring and Fault Diagnosis of A
Low Power Hub Motor Using Feedforward Neural Network,” Computational Intelligence and
Neuroscience, Vol. 2016, 2016.
