A Skin Disease Detection System for Financially Unstable People in Developing Countries

Global Science and Technology Journal Vol. 3. No. 1. March 2015 Issue. Pp. 77 93 A Skin Disease Detection System for Financially Unstable People in Developing Countries Rahat Yasir, Md. Shariful Islam Nibir, and Nova Ahmed Skin diseases are among the most pestiferous and aesthetically displeasing of all maladies. As skin diseases normally take a bit time to be cured and need continuous medicine it is sometimes hard to carry on the treatment specially for the poor people in developing countries like Bangladesh. In this article we propose a method that uses.different types of computer vision based techniques to detect different types of skin diseases based on different types information's collected from patients. As per the proposed method there will be two version of this system one will be a desktop application for algorithm develop and checking, and the other version will be a mobile phone application that will be the handy version of the proposed system. The system will be detecting 9 different skin diseases commonly occurred among the poor people in Bangladesh. Specifically, the covered diseases in this paper are Eczema, Acne, Leprosy, Psoriasis, Scabies, Foot Ulcer, Vitiligo, Tinea Corporis and Pityriasis Rosea. Keywords: Skin Disease, Off the Record, Medical Image processing, Neural Network, Mobile Computing, Telemedicine. 1. Introduction Skin diseases normally occur in human body for several issues. Like scabies occurs in human due to a little bug, acne due to dust and oil glands in the human skin, vitiligo occurs due to fungal attack, foot ulcer for wearing shoes for long time, eczema due to a certain bacteria. Usually most of the skin disease take some time reveal in the skin and need long term treatment. These treatments some time get too expensive to bear for a family in Bangladesh whose monthly income is less than $40 or 3000 BDT per month. A statistics shows that more than 52% of the families in the rural areas of Bangladesh don t have a permanent job they just work as daily workers. In this circumstances spending such money for health problems is impossible for them. Sometimes they go to quack to get some cheap medicine and treatment, which eventually damage their body in the near future. Considering these factors the researchers decided to build a system for the financially unstable people in developing countries. First of all we searched for those skin diseases which are very common in this subcontinent. After doing such studies we found nine diseases that people suffers most and treatment of these diseases might get expensive in some cases. The diseases are Eczema, Acne, Leprosy, Psoriasis, Scabies, Foot Ulcer, Vitiligo, Tinea Corporis and Pityriasis Rosea. At first we went to a well reputed government hospital to collect sample data (image) and user information about these selected diseases. We also conducted a small survey on the patients asking some questions regarding how this type of system will help them, how to they want the system to be developed etc. After getting all these Rahat Yasir and Md. Shariful Islam Nibir, ECE Department, North South University, Bangladesh, Email: rahat.anindo@live.com, nibirsharif@gmail.com Dr. Nova Ahmed, Assistant Professor, ECE Department, North South University, Bangladesh, Email: nova.ahmed.2011@gmail.com

information's we started design the system. We have used eight different types of algorithms for image pre-processing (YcBcr, gray image, sharpening filter, median filter, smooth filter, binary mask, histogram and sobel operator). Our system will take ten different features from image pre-processing results and user inputs (liquid type, liquid colour, elevation, duration, feeling, gender, age). These features are used for training and testing purpose of our feed forward artificial neural networks (ANN). Using artificial neural networks (ANN) as knowledge base appears to be a promising method for diagnosis and possible treatment routines. Even though there have been several researches conducted to detect dermatological skin diseases using Computer Vision based techniques but almost every one worked for only 2-3 diseases. Among them Hashim et al, 2002 Wahab et al, Nasir et al and Shamsul et al have used different types of image pre-processing algorithms for feature extraction purpose. Kabani et al, Nidhal et al, Shamsul et al have used different types of neural networks for dermatological disease detection. Grald et al, Shang et al, Nibaran et al have applied different types of image classification and segmentation algorithms in their proposed method. 2. Related Work Detecting different types of skin diseases from colour image is a very challenging task in computer vision. Finding out different features from the colour skin images of the infected area of different skin diseases and detecting them with a high accuracy rate is the primary purpose of this research. Researchers are working on several algorithms that can be used to detect different types of skin diseases. Hashim et al, 2002 applied different types of matlab tools in images to produce their histograms and colour distribution particularly in the region concerned. Nasir et al compared certain colour space (RGB, CMY, HSV, YCbCr and rgb colour) model of the skin lesions that can be employed to distinguish three types of psoriasis skin diseases infecting the Malaysian population. In the result they saw only RGB colour space is suitable for discriminating psoriasis skin lesions. Shang et al, 2010 compared algorithms of segmentation feature parameter of FCM image segmentation algorithm. Nibaran et al, 2013 compared 4 feature extraction operators with Support Vector Machines (SVM) based classifier. Wahab et al, 2004 presented a texture classification system for three skin diseases (AV, BCC, DLE) and accuracy 78

rate was 70-70-100. Gerald et al, 2006 have introduced an approach that produces accurate overlays of thermal and visual medical images. Chang et al proposed an automatic facial skin defects detection and recognition system. A support-vector-machine (SVM) based classifier is used to classify the potential defects into spot, acne and normal skin and accuracy rate was 98.0%. Hatice et al proposed a software tool diagnosis of erythematosquamous diseases by using the basic and weighted K-NN algorithms on medical data. Kabari et al crated an artificial neural networks system that predict diagnosis and routine treatment for skin diseases patients and their accuracy rate is 90%. M. SHAMSUL et al proposed an automated dermatological diagnostic system. They have used different pre-processing algorithms like ours and used feed forward back propagation artificial neural networks for training and testing purpose. Shuzlina et al elaborates a prototype with back propagation neural network (BPNN) to assist the dermatologist. The use of two feature selection methods namely Correlation Feature Selection (CFS) and Fast Correlation-based Filter (FCBF) help by providing a smaller number of features with greater accuracy and faster response time and accuracy rate was 91.2%. Nidhal et al proposed a psoriais diseases detection system. They used feed forward neural networks to classify input images to be psoriases infected or non psoriasis infected. In this research we are trying to implement similar approach like in order to detect different types of skin diseases from colour image. 3. Architecture and Methodology The system will work on two phases- first pre-process the colour skin images to extract significant features and later identifies the diseases. At first we will be using the colour skin images and then apply 8 different image processing algorithm on it to find some visual pattern and significant features like average colour code of infected area, infected area size in case of pixels and shape or edge detection of an infected area. Then the system will take some user inputs like gender, age, duration, liquid type, liquid colour, elevation and feeling. We will train user input values along with colour skin image extracted features to train and test into a feed forward back propagation artificial neural network to identify the dermatological disease. We divided the whole system into two parts. The first version will be a desktop application, this application will be made to check the algorithms we decided to develop the system. The second version will be the more important one, this will be a mobile phone application which will be a portable system that will be able take photos of the infected area of the skin and take some basic user inputs and the application will instantly take all the inputs from image and user information and will compute and tell which skin disease has occurred. The second version of the system will be moreover a diagnosis system. We are planning to include treatment in the application so that the patients get an overall view of which medicine he/she should go for. Here we have shown the complete architecture of the system below. 79

4. Data Collection Figure 1: Complete Architecture of the system Our first task was to collect necessary data s (image, information) of patients. For fulfilling this purpose we went to Sir Salimullah Medical College and Mitford Hospital, Dhaka, Bangladesh, Department of Dermatology. We took images of total 128 patients of 9 different diseases from the dermatology department. Disease Name Number of Patients Images taken Eczema 7 28 Acne 25 152 Leprosy 5 24 Psoriasis 12 99 Scabies 40 277 Foot Ulcer 6 35 Vitiligo 15 62 Tinea Corporis 10 66 Pityriasis Rosea 8 32 Table 1: Data Collection 80

A specialized doctor was present to validate and record the external data of the patients such as disease history, feeling in diseased part of body, elevation of the diseased region etc. Camera Sony Cyber shot DSC-W550 Focusing The camera was focused manually to adjust the variable nature of natural light Image Resolution 14.1 Megapixels Shutter Speed 1/20 to 1/125 seconds (based on natural lighting) 5. Patients Statistics Table 2: Camera Specification During the data collection process we did three things. 1. Took sample images of each disease 2. Collected basic patient information for each disease 3. Gathered user opinion about our system We collected data of 128 patients from the hospital. We are very thankful to the nurses, stuffs, intern doctors of Sir Salimullah Medical College and Mitford Hospital who helped us enormously during our data collection because the data collection process we went through is not possible without the help of the officials and of course the patients who were so helpful and showed patience in a government hospital and in a country like Bangladesh. Patient Number Male 86 Female 42 Total 128 Table 3: Patient Statistics Skin diseases in a certain area or country depends on its climate and environment. After our whole data collection process we realized that skin diseases based on bacterial and fungal disease occurs in huge number among the poor the people of Bangladesh. The main reason behind this scenario is unhealthy lifestyle they lead. The below graph shows the whole scenario of the diseases among the patients. 81

Patients of Each Disease (per 100) 25 20 15 10 5 0 Category 1 Acne Eczema Scabies Leprocy Vitiligo Psoriasis Foot Ulver Tinea Corporis Pityriasis Rosea Figure 2: Percentage of Patients for Each Disease From the above figure we can see that in every hundred skin disease patients, people suffering from scabies is the highest (20%). Acne and Eczema patients are also very common in Bangladesh (15%). People suffering from Tinea Corporis and Vitiligo are not small in number. Psoriasis and Leprosy patients are not high in number. We found very minimum number of patients of Foot Ulcer and Pityriasis Rosea. 6. Survey We have surveyed with some doctors in Northern International Medical College and Hospital regards to our Skin Disease Detection System. We have conducted our survey in two basic phases. First one is individual and second is group discussion. Ten doctors have participated in our survey. Questions we ve asked during the survey: 1. How do you detect the skin diseases? 2. Do you think is it okay to detect by a mobile application? 3. What is your perception of this kind of system? 4. What do you think about the accuracy of the system? 5. Are the people of rural areas will be benefitted? 6. What do you think about the advantages of primary treatment? 7. Can it will be helpful for early detection? 8. Any suggestion? 82

Figure 2: Percentage of Patients for Each Disease Detection: We have asked several question regarding our Skin Disease Detection System. As our system can detect nine skin diseases, their symptoms and findings may differ in many phases. So, we have asked them the basic question how they detect at the first sight of the disease? They have said, first they listen patient s complaints and observe the symptoms. After analyzing his/her history, they usually go for clinical diagnosis and sign examination. After all the process, they come to a solution and confirm the disease. Dr. Mamum said, First we listen carefully about the patient s problem. Some cases, we detect the disease by observing the infected area, but we won t just confirm without proper diagnosis. Dr. S.M. Jihad said, Different patients have different types of problems. Some have allergy problem not only food but also cold, dust and so on. So, detect and treatment will differ from different perspective Detect by Mobile Application: Another question was, how they think to detect the skin disease by our system. It is typical to detect by using mobile application. Still image sometimes not useful to detect crucial cases. In that case good quality video may help by consulting with doctors. By matching features may good but if not wrong treatment can make a dangerous effect to the patients. 83

Dr. Rashed said, Detecting by a mobile application will be advantageous for everyone, even for the doctors. We can confirm the disease initially by our observation and detecting by mobile application. If the result matches, we can almost sure about the disease. Dr. Mamun said, If it s for doctors, they can easily use it, but patients from remote areas don t have much literacy. They can t use it properly without training. Moreover, taking the picture of the infected area properly is must. Perception: This is really an interesting software. A mobile phone can detect the skin diseases, no can think about it in our country. As skin disease is major issue in medical system. Some people may have skin disease in their secret body parts. They feel shy to visit a doctor. For them, they can easily detect whether they have any skin disease or not. Dr. Onni said, Now a days, mobile phone devices are available to everyone. Almost every person has a mobile device. Nobody likes a skin disease, it destroys people's vivacity. By detecting and taking proper treatment may save from consequences. Accuracy: Accuracy rate should be higher to detect skin diseases. Higher the accuracy, higher the possibility to get proper treatment. Without 100% surety, it won t be wise to start treatment. Dr. Borna said, We need more than 100% surety. After analyzing patients problem, we give blood and urine test to most of the patients. Several factors works behind this. Detecting the disease is really beneficial for both doctors and patients, but patients have to visit a doctor as early as possible to make sure about the disease. Video Calling Can be Helpful for Rural People: Real time communication or video calling and make video database may help to detect diseases. Some extra question about the disease also help to give proper surety. Primary Treatment: For initial diagnosis, it is really helpful. Save time and give a proper suggestion for early treatment. 84

Dr. Koli said, For early detection, there is no doubt it s a wonderful idea. Even we re really interested using this kind of software. It will make good awareness to make an early detection and early treatment. Benefits of Rural People: In rural area, people are not that much aware of using this mobile devices and the detecting software. They would need an assistance to make proper use of this solution. A procedure may also help to the users, about using this software. Some instruction and video record may also be helpful. About its merit and demerits should be included in the information of this solution. Dr. S.M. Jihad said, Yes, they will surely be benefitted. There are lack of doctors and proper treatment facility. Even generating a diagnosis report is a lengthy process for them. Some hospitals don t have the facility of proper equipment. Mobile application will be helpful form them. Helpful for Early Detection: Sometimes, a small skin related problem or common skin diseases may turn in to cancer in future. Skin is very sensitive. We have to take a good care of our skin. Most of the cases, people don t give that much concern about skin related problems. If someone can detect by using this software, it will be a great solution for medical system. Dr. Rashed said, A small skin disease my turn into cancer and we have found many cases where the patients ignored the diseases which turned into skin cancer. Most importantly the patients who have diabetes. They should take care of their skin properly. Suggestion: Skin disease spreads very quickly and contagious. If infection occurs, it turns a real problem for the patients. So early detection within seven days may decrease the probability of unwanted situation even cancer. As we said earlier we gathered some user opinion about our system during data collection period. A total of 73 people participated in this independent survey. Among 52 participants were intern doctors and 21 people were patients suffering from different skin diseases. We collected all responses and generalized them in such manner so that we can see the whole scenario of the society. Our first question was about the general feelings of the participants towards this project. The below graph shows the result 85

User Rating Do You Like Our Solution? 0 10 20 30 40 50 60 70 80 No Yes Figure 3: Survey Response We can see that 95% of the participants liked our solution. The second question was about rating this solution. The below graph shows the result. 5 (Excellent) 27% 1 (Useless) 0% 2 (Not Satisfactory) 3% 3 (Good) 51% 4 (Very Good) 19% 1 (Useless) 2 (Not Satisfactory) 3 (Good) 4 (Very Good) 5 (Excellent) Figure 4: Survey Response Most of the participants appreciated this solution. 37 participants given their response as ''Good'', 14 participants as ''Very good'', and 20 of them said this is ''Excellent''. Third question we asked the participants that which disease among the 9 diseases we are trying to detect will be difficult for our solution. The below graph shows the result. 86

Which Skin Disease you think will be tough to detect using our solution? Tinea Corporis Vitiligo Scabies Leprosy Eczema Psoriasis Acne 0 5 10 15 20 25 30 35 Which Skin Disease you think will be tough to detect using our solution? Figure 5: Survey Response The response of this question differed from person to person. About 40% participants think that Scabies will be difficult to detect, 33% of them voted for Acne. Psoriasis and Eczema will be difficult said 22% of participants for each the disease. 27 % of the participants also thinks Tinea Corporis is the disease. We asked the participants to give their opinion about detecting a disease from some information based on an image and some user inputs in the fourth question. The below graph shows the result. Is it possible to detect skin disease from colour skin image and basic user inputs? 0 10 20 30 40 50 60 70 No Yes Figure 6: Survey Response 88% of the participants gave positive answer and 11 of the participants are not that quite satisfied. The fifth question was about the opinion of the participants if they think the user input we are considering are sufficient or not. The below graph shows the result. 87

What do you think, user inputs that we are using are sufficient to identify the skin disease? 0 10 20 30 40 50 60 No Yes Figure 7: Survey Response We have found some valuable suggestion from our survey participants. Some of the comments they made, "For rational prescription, knowing patient's physical condition, adverse drug reactions, economic status etc. are very important" Detecting and analysing of skin diseases sometimes hard by a mobile application. Some background history can be helpful in this case. "Although a good step and appreciable...but one skin disease can present in multiple form. So detailed personal history, family history, contact history, drug history, typical sites of lesion, elevation is either cystic/solid/infected/blister like. And many more should be include to make the dx more precise. And lastly various presentation of a single lesion have to be present in your apps" "Some more subjects have to be taken care of like family background (whether the disease came from family), the environment where the patient lives" Camera regulation and good lighting system is an important issue, like some of our participants have said, "The app is good, but I think there is a chance of wrong diagnosis if the picture quality is not good & the treatment will not be same 4 all person. Couse choice of drug not only depends on da disease but also on some other factors related 2 the patient. Overall it's a good effort" One of the helpful and appreciating suggestion is, "Add some more detail. There are some diseases that are not considered as skin disease but they show symptoms on skin, try to detect them too. I must say, this is a very good initiative" "I think da app is good, but 88

7. Result In this research 775 colour skin images were used of 128 dermatological disease patients. The proposed system can successfully detect 9 different dermatological diseases with an accuracy of 90%. We have used 15 % of our colour skin images for testing purpose, 10% of images for validation purpose and 75% of images for training purpose. The supervised system works well than semi-supervised and unsupervised system. Detection rate of our supervised system is 90% where the detection rate of semi-supervised system is 88% and 85% for unsupervised system. In the figure-6 we can see different detection rate for 9 different diseases. Detection rate of diseases that has low elevation in the infected area like foot ulcer, vitiligo and psoriasis are very high like 97%, 97% and 91%. Where the detection rate of diseases that has low elevation in the infected area are comparatively low which is between 85-88%. The whole processes of this proposed system starts with taking a colour skin image and pre-process the image by applying 8 different algorithms. Applying pre-processing algorithms we get three image extracted features which we use in our feed forward back-propagation neural network along with seven other user input features. Out of 9 dermatological diseases our system shows best accuracy rate for foot ulcer and vitiligo and our system did not perform according to our expectation for acne and tinea corporis. 100 95 90 85 80 Patients of Each Disease (per 100) 75 Category 1 Acne Eczema Scabies Leprocy Vitiligo Psoriasis Foot Ulver Tinea Corporis Pityriasis Rosea Figure 5: Disease Detection Rate In the Table- 3, we provided sample colour skin images of nine dermatological diseases, approximate elevation rate of each dermatological disease, total image used to train and test our system, successfully disease detected images and accuracy rate for each dermatological disease. We can see that our system performed well for diseases that has low elevation in the infected area. For this type of disease images, system can generate binary image accurately and sobel operator can detect the shape of the infected area more precisely. 89

Table: Detected Disease List with Accuracy Rate 90

From this table we can see that accuracy rate with only user inputs is between 41% to 61% which is not satisfactory. On the other hand accuracy rate with only image processing results is between 61% to 85%. So it shows that our system is well constructed and automated to a great extent. Table: Detected Disease List with/ without Image Processing & User Input 8. Conclusion People of developing countries suffers mostly for different types of skin diseases. Sometime they don t pay heed to any visual changes in their body and at the initial stage they think soon they will be fine. But a simple visual change in the body is the cause of a deadly skin disease. We are developing our system for skin disease patients of developing and under develop countries. We have interviewed many skin disease patients and collect their valuable feedback and our system is the feedback of their valuable feedback. Most of the people liked our solution and the averaging user rating of our solution is more than 3.5. 9. Acknowledgement We are thankful to the Director Brig Dr. Zakir and Dr. Md. Masudul Hoque, Register, Dermatology & Veneriology Department of Sir Salimullah Medical College, Dhaka for helping us to collect Colour skin image data from the Medicine Department of the Hospital. We are also thankful to Adnan Firoze and M.Shamsul Arifin for giving their research data for testing our will be developed system. 91

References: Al-Abbadi, N. K., Dahir, N. S., AL-Dhalimi, M. A., and Restom, H., (2010), 'Psoriasis Detection Using Skin Colour and Texture Features', Journal of Computer Science, vol. 6, no. 6, pp. 626-630 Arifin, M. S., Kibria, M. G., Firoze, A., Amin, M. A., and Yan, H., (2012), 'Dermatological Disease Diagnosis Using Colour-skin Images', Proceedings of the 2012 International Conference on Machine Learning and Cybernetics, Xian, 15-17 July, 2012, IEEE, vol. 5, pp. 1675-1680 Abdul-Rahman, S., Norhan, A. K., Yusoff, M., Mohamed, A., and Mutalib, S., (2012), 'Dermatology Diagnosis with Feature Selection Methods and Artificial Neural Network', 2012 IEEE EMBS International Conference on Biomedical Engineering and Sciences, Langkawi,17-19 December 2012, IEEE, pp. 371 376 Chang, C. Y., and Liao, H. Y., (2011), 'Automatic Facial Skin Defects Detection and Recognition System', 2011 Fifth International Conference on Genetic and Evolutionary Computing, Xiamen, Aug. 29 2011-Sept. 1 2011, IEEE, pp. 260-263 Çataloluk, H., and Kesler, M., (2012), 'A Diagnostic Software Tool for Skin Diseases with Basic and Weighted K-NN'', 2012 IEEE, Innovations in Intelligent Systems and Applications (INISTA), pp. 1-4 Das, N., Pal, A., Mazumder, S., and Sarkar, S., Gangopadhyay, D., and Nasipuri, M., (2013), 'An SVM based skin disease identification using Local Binary Patterns', Third International Conference on Advances in Computing and Communications, Cochin, 29-31 Aug. 2013, IEEE, pp. 208-211 Gao, J., Zhang, J., Fleming, M. G., Pollak, I., and Cognetta, A. B., (1998), 'Segmentation of Dermatoscopic Images by Stabilized Inverse Diffusion Equations', 1998 IEEE, Image Processing, 1998. ICIP 98. Proceedings, vol. 3, pp. 823-827 Hashim, H., Jailani, R., and Taib, M. N., (2002), 'A Visual Record of Medical Skin Disease Imaging Using MATLAB Tools', 2002 Student Conference on Research and Development Proceedings, IEEE, pp. 40-44, Malaysia, 11 April, 2015, http://bit.ly/1elmbpc Happillon, T., Sebiskveradze, D., Vrabie, V., Piot, O., Jeannesson, P., Manfait, M., and Gobinet, C., (2012), 'FCM Parameter Estimation Methods: Application to Infrared Spectral Histology of Human Skin Cancers', 20th European Signal Processing Conference (EUSIPCO 2012) Bucharest, Romania, August 27-31, 2012, IEEE, pp. 1831-1835 Jailani, R., Taib, M. N., and Sulaiman, S., (2003), 'Colour Space for Psoriasis Skin Diseases Analysis', Asia SENSE 2003 SENSOR, pp. 263 268 Kabari, L. G., and Bakpo, F. S., (2009), 'Diagnosing Skin Diseases Using an Artificial Neural Network', IEEE. Adaptive Science & Technology, pp. 187-191 Keke, S., Peng, Z., and Guohui, L., (2010), 'Study on Skin Colour Image Segmentation Used by Fuzzy-c-means arithmetic', 2010 Seventh International Conference on Fuzzy Systems and Knowledge Discovery (FSKD), Yantai, Shandong, 10-12 Aug. 2010, IEEE, vol. 2, pp. 612-615 92

Schaefer, G., Tait, R., and Zhu, S. Y., (2006), 'Overlay of thermal and visual medical images using skin detection and image registration', Proceedings of the 28th IEEE EMBS Annual International Conference, New York City, USA, Aug 30- Sept 3, 2006, IEEE, vol. 1, pp. 965-967, New York, NY, 11 April, 2015, http://bit.ly/1pwg9h9 Schreier, G., Hayn, D., Kastner, P., Koller, S., Salmhofer, W., and Hofmann- Wellenhof, R., (2008), 'A Mobile-Phone based Teledermatology System to support Self-Management of Patients suffering from Psoriasis', 30th Annual International IEEE EMBS Conference Vancouver, August 20-24, 2008, British Columbia, Canada, IEEE, pp. 5338-5341 Singh, H., and Sodhi, J. S., (2013), 'Image Enhancement using Sharpen Filters', International Journal of Latest Trends in Engineering and Technology (IJLTET), vol. 2, issue 2, pp. 84 94 Wahab, N. A., Wahed, M. A., and Mohamed, A. S. A. (2004), ' Texture Features Neural Classifier of Some Skin Diseases', 2004 IEEE, Circuits and Systems, vol. 1, pp. 380-382 93