Skin illnesses and disorders are primarily caused by chemical substances. Extreme temperatures and other physical forces might potentially cause problems. These problems have grown quite widespread, and they must be recognized sooner rather than later in order to be treated effectively. We're attempting to find a solution to this problem as soon as possible in this paper.
Skin disorders affect people not only as a result of external factors, but also as a result of a few diseases (people suffering from cancer or low immunity disorder). Grayscale, K-means clustering, Morphology operation, Color moment, Discrete wavelet transform, and Support Vector Machine classifier are all combined in this study. Kaggle was used for this project. A comparison of different feature extraction techniques and classifiers was carried out. The acquired findings demonstrated the effectiveness of the suggested (SVM) method, as well as how skin illness images were detected from the skin disease dataset.
Table of Contents
I. INTRODUCTION.................................................................................................... 2
II. OVERVIEW OF LITERATURE SURVEY................................................................ 2
III. FLOW CHART:................................................................................................... 3
IV. INPUT IMAGE.................................................................................................... 3
V. PREPROCESSING:.................................................................................................. 3
VI. SEGMENTATION............................................................................................... 3
VII. FEATURE EXTRACTION................................................................................... 4
VIII. IMAGE IDENTIFICATION.................................................................................. 4
IX. RESULT AND ANALYSIS.................................................................................. 4
X. CONCLUSION........................................................................................................ 4
XI. REFERENCES..................................................................................................... 4
Dermatological Disease Detection using Machine Learning – A Review
Angelin Sneha S Student Karunya Institute of Technology and Sciences Coimbatore |
ABSTRACT - Our body's largest organ is our skin. It acts as a shield against mechanical, thermal, and physical harm. It is made up of 3 types of layers, and they are the epidermis, dermis and hypodermis layer. Skin illness is a fairly prevalent condition that my people confront in our day-to-day hurried lives. Skin illnesses and disorders are primarily caused by chemical substances. Extreme temperatures and other physical forces might potentially cause problems. Skin disorders affect people not only as a result of external factors, but also as a result of a few diseases (people suffering from cancer or low immunity disorder). These problems have grown quite widespread, and they must be recognised sooner rather than later in order to be treated effectively. We're attempting to find a solution to this problem as soon as possible in this paper. Grayscale, K-means clustering, Morphology operation, Color moment, Discrete wavelet transform, and Support Vector Machine classifier are all combined in this study. Kaggle was used for this project. Around 202 photos have been gathered (Cancer, Acne, Psoriasis). A comparison of different feature extraction techniques and classifiers was carried out. The acquired findings demonstrated the effectiveness of the suggested (SVM) method, as well as how skin illness images were detected from the skin disease dataset.
Keywords -- skin disease, MATLAB, machine learning, image processing, K-means clustering, colour moment and SVM classifier
I. INTRODUCTION
The skin is the largest organ in the human body. It guards the interior organs against pathogens. It protects bones, muscles, and internal organs by consisting of seven layers of ectodermal tissue. Poor hygiene, rising pollution, global warming, and dangerous UV radiation are all factors that contribute to skin problems. Many skin illnesses can be healed if they are discovered early enough before they spread. It's critical to treat them early on, or else they'll develop into a more serious skin condition. Early detection of numerous skin illnesses (Acne, Cancer, and Psoriasis Dermatitis). Skin disease analysis is a major area of interest in the field of medical image analysis. In order to support dermatologists, Methods based on machine learning are used for detection and classification. The following are the six main steps in this project:
● Input image
● Pre-processing (grayscale is used)
● Segmentation (k-means clustering and morphology operation)
● Feature extraction (discrete wavelet transform and colour moment)
● Classification (SVM classifier)
● Results and analysis.
II. OVERVIEW OF LITERATURE SURVEY
These are reference papers that we used for the literature because they gave us interesting results using a variety of methods and classifiers.
Skin illness is identified using deep learning-based Mobile Net V2 and Long Short-Term Memory. The accuracy of this strategy is greater than 85.34 percentand the database was collected from HAM1000(KAGGLE) [1]. Skin disease is detected via computer-aided diagnostics (CAD). CNN models that are currently state-of-the-art can outperform models that have been established in the past. The database was collected from DermNet. Pre-processing, self-supervised learning, transfer learning, and customised CNN architecture approaches were employed, resulting in a high degree of accuracy [2]. A standard hybrid framework for detecting skin disorders is presented in this paper. The images are collected from different websites. With a maximum system accuracy of around 97 percent, the uniform distribution-based segmentation combined with an active contour method is resilient. Furthermore, with an accuracy of around 98 percent, DWT using the PCA approach is more efficient than other methods for feature extraction. Furthermore, the SVM with KNN approach is more resilient than other methods for classification, with an accuracy of around 98 percent [3]. Based on image processing and machine learning approaches, this research provides a method for detecting skin diseases. The proposed approach is especially useful in rural areas where dermatologists are rare. They employ a PyCharm-based python script for the experimental outcomes in this proposed system and the database was collected from different websites [4]. Melanoma, eczema, and impetigo are three different types of skin illnesses which are collected from different websites that are detected using machine learning and image processing in this paper. This demonstrates a great level of precision [5]. The database was collected from various websites. To detect the disorders, image processing techniques from the specialised field of Medical Science that deals with the diagnosis and treatment of skin diseases were utilised. This project relied heavily on the Octave tool [6]. From ISIC, OLE, DermNet, PH2, Dermis, DermQuest database was collected. According to this study, the diagnostic accuracy of image processing algorithms ranged from 50% to 100%. Tissue characteristics were treated with a precision of 94 percent or higher [7]. The classification of skin illness in this research is based on 938 pictures. The classification of skin condition is done with a 70 percent accuracy using CNN algorithms, and they also attempted Alex Net, which has an accuracy of 80 percent [8]. The classification of skin illnesses in humans into four classes is determined in this work, including Benign Keratosis, Melanoma, Nevus, and Vascular which are collected from different websites. The segmentation approach was K-Means Clustering, and the feature extraction method was Discrete Wavelet Transform (DWT) and Color Moments feature extraction. It has been shown to be more effective than traditional biopsy procedures [9]. From International Skin Imaging Collaboration, a database was collected and to detect different types of skin illness, a noval skin image segmentation technique was presented with the help of a naïve bayes classifier. The accuracy rate for benign cases is 94.3 percent, 91.2 percent for melanoma, and 93.9 percent for keratosis [10]. A method for detecting skin illness based on image processing has been proposed. The database was collected from online websites. Three diseases were successfully recognised using this technology, with an accuracy rate of 100% [11]. From various websites database was collected and Image processing techniques such as adaptive thresholding, edge detection, k-means clustering, and morphology-based image segmentation are utilised to detect four types of skin disorders in this research, and four different segmentation algorithms are applied for each of the four skin diseases [12]. The database was collected from different websites and using the KK classifier, a colour phase model was created for detecting and classifying various types of skin illness. The HSV colour phase model has 91.80% accuracy, whereas the lab colour phase model has 81.60 percent accuracy [13]. In this study, five different machine learning algorithms were chosen and used to detect different forms of skin disease: Random Forest, naive bayes, logistic regression, kernel SVM, and CNN [14]. Three forms of skin disease which were collected from the DermNet and DermWed were detected using a method based on vertical image segmentation [15]. They create a system that uses machine learning to identify different types of skin illness in this research. The database was collected from G.S.M.C. KEM hospital, Parel, Mumbai was used. A level of accuracy of above 90% was achieved [16]. From ISIC 2018, the database was collected and used in this paper. The Variational Autoencoder was utilised in this study to detect skin disorders with a 0.779 AUCROC. This model can detect melanoma with an AUC ROC of 0.864 and actinic keratosis with an AUCROC of 0.872 [17]. Using a new detection approach, three types of skin illness can be diagnosed which were collected from the Sino medical websites. They used 20 test samples and 10 standard samples of each disease to identify it. The accuracy rate in this publication is around 90% [18]. Deep learning and other computer vision-based approaches are being used to predict different types of skin illness. Twenty diseases were collected from different websites and were predicted with an accuracy of 88 percent using deep learning technology [19]. The NN-NSGA-II model was used to categorise two different skin disorders, Based Cell Carcinoma and Skin Angioma which was collected from international skin imaging collaboration (ISIC), in this paper [20]. The meta-heuristic supported artificial neural network technique is employed in this article, and the major skin illnesses used in detection are angioma, basal cell carcinoma network, and lentigo simplex which was downloaded from the international skin imaging collaboration (ISIC). In this paper, a computer-assisted automatic method for the classification of skin lesions is provided, which uses an NSGA II trained neural network. This method has an accuracy of 87.92 percent, which is higher than previous methods [21]. Five distinct machine learning methods, including Random Forest, Naive Bayes, Logistic Regression, Kernel SVM, and CNN, were chosen and used to diagnose various types of skin illness which was downloaded from the DermNet websites in this work [22]. The technology proposed in this paper is based on image processing techniques and is mobile. Image processing is used to diagnose skin disease, and the database collected from different websites online contains six prevalent skin diseases [23]. Skin illnesses such as psoriasis, warts, moles, and eczema were detected with an accuracy of over 90% using texture-based characteristics generated from the GLCM matrix [24]. The two forms of skin illnesses (acne and psoriasis) are detected using KNN, and the Samone parameter values are produced using the active contour approach. The detection technology is cost-effective and gets the job done quickly [25]. They adopt a two-stage strategy to identify the disease in this paper, combining computer vision and machine learning. When it came to detecting skin problems, 95 percent accuracy was achieved [26]. The dataset consists of 775 skin photos for 9 diseases, and the method uses computer vision-based algorithms to diagnose various types of skin disease. This technique has been successful in diagnosing 9 distinct diseases with a 90% accuracy rate [27]. This research outlines the creation of a low-cost smartphone-based intelligent system with integrated cameras for skin disease detection. The skin photos were analysed using an artificial neural network technique [28].
III. FLOW CHART:
[This illustration is not included in this text sample]
IV. INPUT IMAGE
A dermatoscopic image of skin is used as the input image. Nearly 202 images of skin diseases were downloaded for free from the Kaggle online website and classified into three categories, as shown in figures 1, 2, and 3.
Editor´s note: These images were removed for copyright reasons.
V. PREPROCESSING:
Pre-processing is the first stage of image restoration, removing unwanted parts in the background of skin images as well as removing noise. The input data is transformed into a suitable data form with the required processing during the preparation stage. This method was developed to convert an RGB image to a grayscale image. It is used to simplify the image model, which is then further simplified during the segmentation process, where even noises can be detected. The grayscale images are produced after pre-processing the original images.
VI. SEGMENTATION
Image segmentation is the most critical step in recognising an image. The basic objective of image segmentation is to separate images into various regions and to judge where in the image more focus should be placed in comparison to the backdrop. An accurate+ segmentation of skin images can aid in the diagnosis by accurately defining the affected skin region from the input image. To segment the skin illness image, we apply the k-means clustering method. K-means clustering divides the given image data into numerous clusters, or regions, based on the distance between them. Between the image data and each cluster's centroid. Each time, a new set of data points is considered, and the distances are calculated. The set of records is compared, and the set of data points resulting in the shortest distance is chosen. We used K-means clustering to segment the image, and the number of K is 3.
VII. FEATURE EXTRACTION
Feature extraction is a procedure in which the raw data is further condensed into manageable groupings for processing by reducing the dimension. Skin illness can be distinguished by a variety of characteristics such as texture and colour. Discrete wavelet Transform and Color Feature are the two main forms of extraction suggested.
- Discrete Wavelet Transform
A discrete wavelet transform is a TransForce that decomposes a signal into a number of sets, each set containing a time series of coefficients that describe the signal's time evolution. The textural features in skin disease are retrieved using the discrete wavelet transform.
The discrete wavelet transform is one way for extracting information from an image. Decomposition is performed on the image in DWT. A high pass filter is used for high frequency transmissions, whereas a low pass filter is used for low frequency signals. The original image is then deconstructed and divided into four components. Each sub image is 1/4 times larger than the original.
They appear to be a simplified version of the original image, with a smoother appearance due to the presence of a low frequency component. The wavelet level will be used to guide the decomposition process.
- Color Moment
Color moments, like central moments, are metrics that characterise the colour distribution of an image. As a colour moment, the colour feature extraction approach was developed. It will treat an image's distribution as a probability distribution. The colour moment approach was shown to be the most accurate in identifying the colour aspects of skin illness. The premise that the colour distribution of an image can be described as a probability distribution is the foundation of this approach.
VIII. IMAGE IDENTIFICATION
This is the final stage, which is separated into two parts and involves the detection and classification of skin disease images by kind. The stage is broken into two parts: training and testing. The training process is where the data is trained and new data is processed. The outcome of classification is tested with the help of the testing process, which uses new data obtained from the training phase. Support Vector Machine (SVM) is the method that is proposed at this level.
The Support Vector Machine is a supervised learning system that can be applied to classification and regression problems. SVMs are supervised learning algorithms that can be used for classification, regression, and outlier detection. SVM's benefits are effective in high-dimensional spaces. The quantity of samples is still more effective. SVM's classification concept creates a hyperplane that serves as a divider between two sets of data. The hyperplane will separate two independent data sets by measuring the margin line and looking at a maximum point. The margin is the distance between the hyperplane and the nearest data. A support vector is a set of data that is the most similar. A support vector machine is the method that is proposed at this level.
IX. RESULT AND ANALYSIS
Early detection of skin illness has an important role in lowering the mortality rate. Image segmentation using machine segmentation is one of the most effective methods for early identification and treatment of skin disease. This paper has gone over every skin disease detection approach in detail. Image segmentation accuracy was measured using several criteria, including sensitivity, specificity, and accuracy. When compared to other ways, many of the methods produced positive results. MATLAB is used to implement the system. The input image was first pre-processed, and the pre-processed data was then segmented using k-means clustering, with the segmented image shown in figures 10, 11 and 12.
X. CONCLUSION
Skin diseases are becoming more common, and diagnosing them is both costly and time consuming. Any inaccurate diagnosis that leads to inefficient therapy puts the patient's health in threat. This research analysed the literature on skin disease detection and identified the methods that must be followed to detect skin diseases. Each phase explains the approaches and strategies that are useful in the process. As a result, using machine learning to diagnose skin illnesses improves treatment and the patient's chances of recovery in this study. The presented strategy is quite useful and can be applied in real-world scenarios. It has the ability to help people from all over the world. It might also be user-friendly, making it accessible to everybody. Machine learning is intended to aid dermatologists in doing real-time diagnostic tests. This technique can aid new medical practitioners in appropriately recognising the ailment in the event of a misdiagnosis.
XI. REFERENCES
1) Parvathaneni Naga Srinivasu, Jalluri Gnana SivaSai, Muhammed Fazal Ijaz, Akash Kumar Bhoi, Wonjoon KIm and James Jin Kang, MDPI publisher of open access journal, “Classification of Skin Disease Using Deep Learning Neural Networks with Mobile Net V2 and LSTM” 2021
2) Ha Min Son, Wooho Jeon, Jinhyun kim, chan yeong Heo, Hye Jin Yoon, Ji-Ung Park and Tai-Myoung Chung, Scientific Reports, “AI‑based localization and classification of skin disease with erythema” 2021
3) Md. AlMamun and Mohammad Shorif Uddin, Journal of Computer and Communications, “Hybrid Methodologies for Segmentation and Classification of Skin Diseases: A Study” 2021
4) Leelavathy S, Jaichandran R, Shobana R, Vasudevan, Sreejith S Prasad and Nihad, European Journal of Molecular & Clinical Medicine, “Skin Disease Detection using computer vision and Machine learning technique” 2020, Volume 7, Issue 4,
5) Divya Shree D V, Goggi Pragna, Prathibha L, Sushmetha G R and Mary M'Dsouza, International Journal of progressive Research in Science and Engineering, “Skin disease detection using image processing and neural networks” July 2020, Volume 1, Issue 4
6) Anita Thakur, A.J. Singh, International Journal of Scientific and Technology Research “Feature Analysis of skin Disease Images Using Various Techniques of Digital Image Processing”, 2020, Vol.9, Issue;03
7) Saja Salim Mohammed and Mustafa Al-Tuwaijari, 2nd International Scientific Conference of Engineering Sciences, “Skin Disease Classification System Based on Machine Learning Technique: A Survey” 2020
8) S. Malliga, G. Sherly Infanta, S. Sindoora and S. Yogarasi, International Journal of Advanced Science and Technology “Skin Disease Detection and Classification using Deep Learning Algorithms” 2020
9) I Ketut Gede Darma Putra, Ni Putu Ayu Oka Wiastini, Kadek Suar Wibawa and I made Suwija Putra, International Journal of Machine Learning and Computing “Identification of skin disease using K-Means Clustering, Discrete Wavelet Transform, Colour Moments and Support Vector Machine” Vol.10, No.5, September 2020
10) V. R. Balaji, S. T. Suganthi, R. Raja Devi, V. Krishna Kumar, B. Saravana Balaji and Sanjeevi Pandiyan, ScienceDirect, “Skin disease detection and segmentation using dynamic graph cut algorithm and classification through Naive Bayes classifier” 2020
11) Nawal soliman ALKolifi ALEnezi, 16th International Learning & Technology Conference, “A Method of Skin Disease Detection Using Image Processing and Machine Learning” 2019
12) Kyamelia Roy, Sheli Sinha Chaudhuri, Sanjana Ghosh, Swarna Kamal Dutta, Proggya Chakraborty and Rudradeep Sarkar, International Conference on Opto-Electronics and Applied Optics (IEM OPTRONIX), “Skin Disease detection based on different Segmentation Techniques” 2019
13) A. V. Ubale and P. L. Paikrao, International Research Journal of Engineering and Technology (IRJET), “Detection and Classification of Skin Disease using Different Color Phase Models” 2019, Vol.6, Issue 7
14) Shuchi Bhadula, Sachine Sharma, Piyush Juyal, Chitransh Kulshrestha, International Journal of Innovative Technology and Exploring Engineering (IJITEE), “Machine learning algorithms-based skin disease detection” 2019, Vol.9, Issue 2
15) Ismail Saif Sulaiman Al Shabibi and Sreedevi Koottala, Journal of Student Research, “Detection of Skin Diseases Using MATLAB” 2019
16) Ms. Seema Kolkur, Dr. D. R. Kalbande and Dr. Vidya Kharkar, International Journal of Computational Intelligence Research “Machine Learning Approaches to Multi-Class Human Skin Disease Detection”, 2018, Vol.14, No.1
17) Yuchen Lu and Peng Xu, the 8th International Symposium on Intelligent Systems Techniques for Ad hoc and Wireless Sensor Networks (IST-AWSN), “Anomaly Detection for Skin Disease Images Using Variational Autoencoder” 2018
18) Li-sheng Wei, Quan Gan and Tao Ji, Computational and Mathematical Methods in Medicine, “Skin Disease Recognition Method Based on Image Color and Texture Features” 2018
19) Sourav Kumar Patnaik, Mansher Singh Sidhu, Yaagyanika Gehlot, Bhairvi Sharma and P Muthu, Biomedical & Pharmacology Journal, “Automated Skin Disease Identification using Deep Learning Algorithm” September 2018, volume 11(3)
20) Shouvik Chakraborty, Kalyani Mali, Sankhadeep Chatterjee, Sumit Anand, Aavery Basu, Soumen Banerjee, Mitali Das and Abhishek Bhattacharya, IEEE 8th Annual Ubiquitous Computing, Electronics and Mobile Communication Conference (UEMCON), “Image Based Skin Disease Detection using Hybrid Neural Network Coupled Bag-of-Features” 2017
21) Shouvik Chakraborty, Kalyani Mali, Sankhadeep Chatterjee, Soumen Banerjee, Kaustav Guha Mazumdar, Mainak Debnath, Pikorab Basu, Soumyadip Bose and Kyamelia Roy, Annual Industrial Automation and Electromechanical Engineering Conference (IEMECON), “Detection of Skin disease using Metaheuristic supported Artificial Neural Networks” 2017
22) Md.Nazrul Islam, Jaime Gallardo-Alvarado, Masyitah Abu, Nurfazlyna Aneem Salman, Sahanah a/p Pandu Rengan and sakina Said, IEEE 8th Control and System Graduate Research Colloquium, “Skin Disease Recognition Using Texture Analysis” 2017
23) Archana Ajith, Vrinda Goel, Priyanka Vazirani and Dr. M. Mani Roja, International Conference on Intelligent Computing and Control Systems, “Digital Dermatology Skin Disease Detection Model using Image Processing” 2017
24) Seema Kolkur and D. R. Kalbande, International Conference on ICT in Business Industry & Government (ICTBIG), “Survey of texture-based feature extraction for skin disease detection” 2016
25) Manisk Kumar and Rajiv Kumar, ARPN Journal of Engineering and Applied Sciences, “An intelligent system to diagnosis the skin disease” 2016
26) Vinayshekhar Bannihatti Kumar, Sujay S Kumar and Varun Saboo, IEEE Third International Conference on Artificial Intelligence and Pattern Recognition (AIPR), “Dermatological disease detection using image processing and machine learning” 2016
27) Rahat Yasir, Md. Ashiqur Rahman and Nova Ahmed, 8th International Conference on Electrical and Computer Engineering, “Dermatological Disease Detection using Image Processing and Artificial Neural Network” 2014
28) Abderrahim Bourouis, Ali Zerdazi, Mohammed Feham and Abdelhamid Bouchachia, the 8th International Symposium on Intelligent Systems Techniques for Ad hoc and Wireless Sensor Networks (IST-AWSN), “M-Health: Skin Disease Analysis System Using Smartphone's Camera”