Mobilebased Intelligent Skin Diseases Diagnosis System Subtitle

Mobile-based Intelligent Skin Diseases Diagnosis System

Subtitle as needed (paper subtitle)

Abdulfattah Esmail Hasan Abdullah Ba Alawi

Software Engineering. Faculty of Engineering & IT

Taiz University

Taiz, Yemen

Dr. Amer Ali Sallam

Computer Engineering. Faculty of Engineering & IT

Taiz University

Taiz, Yemen

Abstract Skin diseases are the most common diseases in humans. The inherent variability in the appearance of skin diseases makes it hard even for medical experts to detect disease’s type from dermoscopic images.

Recent advances in image processing using the Convolution Neural Networks have led to better results in diagnosing systems. We aim to develop an advanced diagnosis system in a manner that meets the requirements of real time and extensibility for medical services for skin diseases detection that provides offline detection for the users who have not connection to Internet and online diagnosing service and communication service with a medical team. First, the user capture the affected area and get offline immediate report by on-device diagnosis.

Get quality help now
Verified writer

Proficient in: Cancer

4.9 (247)

“ Rhizman is absolutely amazing at what he does . I highly recommend him if you need an assignment done ”

+84 relevant experts are online
Hire writer

In addition to online accurate diagnosis option using on-cloud service. The third service is an option to communicate with dermatologists and get medical recommendation. New images that labelled by dermatologists used to retrain the model to enhance system accuracy. As a reason for maximizing the number of users, the system implemented on a mobile-based environment, by the growing numbers of portable apps allows it easy for people to obtain up-to-date data. Users are familiar to looking for answers from the virtual globe, including health issues.

Get to Know The Price Estimate For Your Paper
Number of pages
Email Invalid email

By clicking “Check Writers’ Offers”, you agree to our terms of service and privacy policy. We’ll occasionally send you promo and account related email

"You must agree to out terms of services and privacy policy"
Check writers' offers

You won’t be charged yet!

The following experimental results demonstrate the e?ectiveness and feasibility of the proposed method. Test accuracy is 95.6% for benign cases, for the melanoma case it is 90.78 %.

Keywords Deep Convolution Neural Networks; Virtual globe; dermoscopic; skin lesion;


Skin disease is the most common disease in the globe. It is known as a pathological condition on the body’s surface [1]. It has various appearances and various degree of affects, from slight affect such as changing body characteristics to impact affect like death. Diagnosing skin diseases in early stages is very important due to a high survival possibility especially for skin cancer. That account for around 80% of all newly diagnosed cancers. Early diagnosis of melanoma has an elevated cure rate and a relative survival rate of 99% for 5 years. However, since non-melanoma skin cancer can easily spread to other areas of the body, the comparative 5-year mortality ratio in the long-term falls to 20% [1] or 18% [23]. What show the wide spread of this skin diseases according to American Cancer Society report [4], an estimated 96,480 new cases of melanoma will be diagnosed in 2019. The reason for using mobile-based system no other platform, in the report [5], the number of users of smartphones is expected to reach 7.2 billion in 2024. General diagnostic services can be provided at low costs [6]. Therefore, it is difficult to be classified easily. Therefore, we suggested a method using mobile-based vision techniques to detect different types of skin diseases. In the paper [3] introduced mobile devices that have deep neural network could possibly extend the variety of dermatologists outside the outpatient department. The traditional systems for detecting skin disease complete the classification production by extracting picture information as input features. The research that is currently carried out adopts the deeper framework for automating teaching of features[7-9], with the priority of precision of automatic classification depending on pathological skin dataset being acquired. Meanwhile, the present system is a centralized system that needs an active expert update to include a static and centralized database that limits user mobility and cannot conduct a convenient and effective self-check. Furthermore, the centralized system cannot provide enough resources to support the individualized database for different population groups and cannot make a good judgment of paroxysmal diseases because of centralizing the database [1]. According to the latest World Health Organization (WHO) data published in 2017, skin disease deaths in Yemen reached 166 or 0.11% of total deaths. The age-adjusted death rate is 1.46 per 100,000 of population ranks Yemen #100 in the world [39]. The arrival of smartphones into many fields, for example medical fields, has resulted to the development of new technology to help individuals to identify and diagnose illness precisely and credibly where the doctor’s knowledge, in relation to software precision, leads to exposure to high-patient confidence and saves human life.

While skin disease is a global problem, there are many researches in this field. Existed artificial skin diseases diagnosis systems have few solutions available, which are still under research developments. Certain limitations and drawbacks are identified in those hence this work tries overcome the existing problems with different approach [16]. In recent years, image processing has performed an important part in this study field and has commonly used to detect skin diseases. These illnesses can be recognized accurately with combined techniques like image processing, data mining and machine learning algorithms etc.

In 2012, Shervan et al [24] used a method of Color-Based Image Retrieval (CBIR) to detect the types of skin with high accuracy. First, skin pictures were trained during training phase, and then the skin region from non-skin regions was identified during test phase. With finding means of CBIR technique and picture tiling, and then discovering the connection between pixels and its neighbors. Okuboyejo et al [17] used image classification in an automated skin disease diagnosis, which collects the images outcomes from the past Pigmented Skin Injury (PSL) and comments by medical professionals. Part of the system is used to evaluate, store and ranks information in the Image Storage using computational intelligent techniques depending on texture and potentially morphological pictures. The system used only to classify melanoma and nevi using pattern recognition algorithms. In 2014, Munirah et al [18] developed an online diagnostic system for children with skin diseases. This system allows users to identify and provide advice for kids’ treatment online. The customer answers requests based on his or her skin disease or diseases. It helps users to identify skin diseases in kids via the online system and gives users a helpful recommendation. The system was created to assist users diagnose and detect skin diseases for kids and offer suggestive therapy quicker and more conveniently through the Internet. In addition, this system also helps to prevent kids from highly affected by skin diseases by offering relatives with adequate therapy data when using the system. Although the Diagnosis System has effectively been established for children online, it can still be improved with more illnesses, more conditions and images for skin diseases or to cope with other age groups such as adults. In the same year, Yasir et al [19] introduced a system for detection different types of skin conditions. The system examines a picture of human body affected and detects the condition at nearly a 90% accuracy. It detects limited number of diseases almost nine distinct skin conditions. Eczema, Acne, Leprosy, Scabies, Ulcer Foot, Tinea Corporis and Rosea Pitriasis. In the study of 128 dermatology participants, 775 color pictures of the skin were used. Nine distinct dermatological illnesses can be effectively detected with a precise 90% by the suggested system. They used 15% of the color skin images for test purposes, 10% for testing purposes and 75% for training purposes. Detection rate while using supervised algorithms was 90% where the detection rate of semi-supervised system is 88% and 85% for unsupervised system. Vividly, this system needs to enhance accuracy and implement this system using android to maximize the number of users. Kabari et al [10] developed a system with Artificial Neural Network to predict diagnosis of skin diseases and routine treatments for patients with a 90% accuracy rate. M. SHAMSUL et al [11] introduced an automated dermatological diagnostic system based on used various pre-processing algorithms such as artificial neural networks for training and testing purpose. Chang et al [15] suggested an automatic system for the detection and recognition of facial skin defects with 98.0% accuracy rate.

In the year 2015, there were many attempts for developing better skin diseases diagnosis systems, for example, Amarathunga et al [20] introduced a diagnostic system for skin diseases that enables users to recognize individual skin disorders and to offer consultation or medical treatment. The process was done by image processing operations flowed by data mining techniques to identify skin disease. The accuracy of eczema disease identification is 85%, Impetigo is 95 percent and Melanoma is 85%. The limited number of skin diseases, which are supported by this system (three diseases only), make a need to expand it. In a filed of mobile vision, Aruta et al [21] introduced a mobile medical assistance for the diagnosis of skin diseases through Case-Based Reasoning (CBR). That is used to assist users pre-examine the condition of their skin whether or not they have an illness. The achieved accuracy of eczema disease identification is 85 percent, Impetigo is 95 percent and Melanoma is 85 percent. The methodology employed in that study is CBR that is used for establishing a new knowledge base and the image processing technology to determine the symptoms base of an individual’s skin problem in the newly captured image. Only six diseases can successfully identified by this system with an accuracy of 90%. In 2016, P. S. Ambad et al [36] presented an Image Analysis System to detect skin diseases. The image analysis method in which users can bring pictures of skin of distinct mole type or rashes. System process is to analyze pictures that provide users with medical assistance and includes automatic avoidance and identification of skin diseases include skin cancer, psoriasis cases. The dataset used by this system contains 130 images of dermatologic diseases and reaches 90% accuracy rate.

In 2017, D.S. Zingade et al [22] designed a system that detect skin diseases using ANN. It helps to diagnose the illness with providing the image of the area of the skin affected by skin diseases. Recognizing system utilizes characteristics obtained from the body image through the handling algorithm together with back propagation artificial neural network and image processing. A three-stage system that operates. The first stage involves handling the picture of the skins associated with illnesses to acquire significant characteristics, such as Region Of Interest ( ROI ), second stage consists of the training stage to train the neural network to identify the dermatological illnesses. The third stage includes feed forward disease detection by back propagation neural network.

In 2018, N. Singh et al [23] outlined a study into computer-aided diagnostic techniques for melanoma risk assessment and premature testing. In this approach J. Rathod et al [25], discussed an automated system for recognition of skin diseases using CNN algorithm for feature extraction. The image of affected area is classified using CNN and softmax classifier. Five diseases were initially tested with accuracy of 70%. The images of the dataset are available on Dermnet ( The accuracy can be enhanced to reach 90% using a large dataset for training. Moreover, Zulfikar et al [26] discussed an Android-based application, which analyses skin images using the canny edge detection. After uploading two images, the app analyzes the first image while the second is taken as a reference to comparing process. The system analyze the unhealthy image using mean value, which states the differences degree between the two images. Then, mean value used in decision-making that generate detection result. The system was built with an OpenCV library for implementing the canny edge detection. In 2018, Wei et al [27] developed a diagnosis of skin diseases. The system could identify three types of skin diseases, which are herpes, dermatitis and psoriasis. Gray-Level Co-occurrence Matrix (GLCM) is used as a new identification method to enhance diagnostic accuracy. GLCM used for extracting texture features.

In May 2019, E. Akar et al [28] introduced skin lesion diagnosis system based on cloud using CNN. The user take a photo using android-based application and upload the image to cloud. Then, a deep learning based classifier hosted in server to filter and classify uploaded lesion image. ISIC 2018 dataset and Caltech 101 dataset were used to train a model. Caltech is for non-lesion and the ISIC dataset for lesion images. To improve accuracy of the model, more data would need to be gathered. The diagnosis CNN was trained using transfer learning with roughly 3,000 images and that is not nearly enough to allow experimentation with diagnosis CNN, whose accuracy is near dermatologist level. Sreelatha et al [29] introduced an intelligent system for diagnosing melanoma skin cancer in early stages with dermoscopic images using Gradient and Feature Adaptive Contour (GFAC) model. However, this app is still not perfect and need more enhancement. There are several things can be done to improve the quality of the analytical desires various classification algorithms like SVM, AdaBoost and Bag of Features (BoF) is utilized to determine the efficiency of the image segmentation technique. Wenlitu et al [30], introduced a novel technique using Dense-Residual (Decoder – encoder) layers to increase segmentation accuracy. It uses five (Dense – Residual) layers to achieve accurate segmentation of skin lesion. The goal is to separate the lesion in the skin image from the surrounding normal tissue without manual intervention. PH2 dataset and ISBI 2017 dataset are used. Ech-Cherif et al [31] discussed a deep neural network based mobile dermoscopic application for triaging skin cancer detection to classify suspicious lesions. They used a dataset of 48,373 dermoscopic images collected from three different archives labelled and validated by expert dermatologists. They used a trained CNN to binary classification of skin lesions into benign or malignant classes. Using batch size of 32, the overall accuracy was 91.33%. Moreover, Barata et al [32] discussed a hierarchical skin lesion organization, to develop a deep learning system with a structured classification to classify suspicious lesions. Moreover, underline the advantages of structured approaches to dermoscopic image classification. The overall accuracy reaches 91.33%. The model is evaluated on the publicly available ISIC 2017 dataset. This set contains 2,750 images each belonging to one of the following three classes: melanoma and nevus, both melanocytic, and seborrheic keratosis (non-melanocytic). Moreover, the obtained attention maps show that the system is able to identify clinically relevant regions in the lesions, as well as to provide more insightful information on the importance of the different image regions in the diagnosis. Furthermore, Astorino et al [33] introduced Melanoma Detection by Means of Multiple Instance Learning. Interdisciplinary Sciences. An approach of Multiple Instance Learning (MIL), the object (MIL) called bag and its items (positive and negative items) are called instances. they used an MIL algorithm on certain clinical information made up of colored dermoscopic images in order to distinguish melanomas (positive instance) from prevalent nevi (negative instance). Their results was courageous, as followed: accuracy is 92.50%, sensitivity is 97.50% and specificity is 87.50%. In the smart devices side, A. Hameed et al [34] presented skin diseases detection with android platform, offering true and helpful dermatologic data regarding four skin diseases like acne, skin hurry, and Melanoma. The request can bring and email your dermatologist with a commentary and submit both clip and ordinary and magnified photographs. It is a stepping towards making mobile application helping to diagnose and cure their users without teledermatology visits. The dataset includes 100 image of each class, 50 of them were colored images, and another 50 were gray images. The dataset was segmented to 60 percent of images used for training the model and 40 percent used for evaluation. This system accuracy is about 92%. Recently, an intelligent system was introduced by S. U. Ahmed et al [35] for detecting skin diseases using machine learning algorithms, in addition to provide a communicating services with doctor, the system implemented on Android Studio and linked with Raspberry Pi that hosts one of Google’s deep learning library called Tensorflow. Convolution neural Network implemented, and was useful to identify four types of skin diseases, which are Melanoma, Acne, Cystic and Eczema. Therefore, the user can upload the image of the infected area with disorder’s conditions. Then e-doctor can identify the diseases and make appointment in emergency cases. This system was able to detect four types of skin diseases. In addition, Min Chen et al [1] introduced an intelligence system called AI-Skin. Skin disease recognition based on self-learning and wide data collection through a closed-loop framework use accumulated database that store historical uploaded images from user to enhance detection accuracy of various groups. A structure for medical AI depending on evolution of information length and self-learning. The dermatologists were invited to classify datasets images, which reaches 6144 images. The labels of images contain 14 classes, which are facial acnes, nevus, large pores, forehead acnes, alar acnes, acne marks, chloasma, dark circles, blackheads, pregnant spots, age spots, radiation spots, sunburn spots, and wrinkles. The universality of the extended algorithm interface has been verified using the three cloud-trained learning models, i.e. LeNet-5, AlexNet and VGG16. However, the accurate model was AlexNet that has the accuracy when they use five classes of skin diseases, 0.79 for Skin acnes, 0.80 for Skin spots, 0.91 for Skin blackheads, 0.78 for dark circles, and 0.95 for clean face. With that system, the high- and low-resolution of the uploaded image does not affect the classification much.

Because of the problems that face skin doctors in detecting skin diseases, those problems such as need time in detection process, Lack of awareness of all cases of skin disease and the some erroneous diagnoses of skin disease. In other words, affected people need time, cost, and efforts to come to hospital for diagnosis purposes. Therefore, we present an intelligent system for diagnosing skin diseases and provide accurate medical report with offering medical recommendation. The system will help users diagnosing their skin disorders easily using their smartphones regardless their places in timely manner.

This paper consist of Background on skin diseases diagnosis, the problem, and to where researchers stepped and current deep learning based diagnosis solutions are described in Sect. II. In Sect. III, the architecture and the methodology of the system and its components are explained. In Sects. IV and V, the training, and results of the deep learning based diagnosis and preliminary classifiers are discussed. The design and implementation of the mobile application demo is presented in Sect. VI, Future work is discussed in Sect. VII.

Similar topics:

Medicine Topics

Cite this page

Mobilebased Intelligent Skin Diseases Diagnosis System Subtitle. (2019, Dec 20). Retrieved from

👋 Hi! I’m your smart assistant Amy!

Don’t know where to start? Type your requirements and I’ll connect you to an academic expert within 3 minutes.

get help with your assignment