2168-2194 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JBHI.2022.3149288, IEEE Journal of Biomedical and Health Informatics IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS, VOL. XX, NO. XX, XXXX 2022 1 Federated Machine Learning for Detection of Skin Diseases and Enhancement of Internet of Medical Things (IoMT) Security Md. Nazmul Hossen, Student Member, IEEE , Vijayakumari Panneerselvam, Deepika Koundal, Kawsar Ahmed, Graduate Student Member, IEEE , Francis M. Bui, Member, IEEE , and Sobhy M. Ibrahim Abstract — Human skin disease, the most infectious dermatolog- ical ailment globally, is initially diagnosed by sight. Some clinical screening and dermoscopic analysis of skin biopsies and scrap- ings for accurate classification are medically compulsory. Classifi- cation of skin diseases using medical images is more challenging because of the complex formation and variant colors of the disease and data security concerns. Both the Convolution Neural Network (CNN) for classification and a federated learning approach for data privacy preservation show staggering performance in the realm of medical imaging fields. In this paper, a custom image dataset was prepared with four classes of skin disease, a CNN model was suggested and compared with several benchmark CNN algorithms, and an experiment was carried out to ensure data privacy using a federated learning approach. An image augmentation strategy was followed to enlarge the dataset and make the model more general. The proposed model achieved a precision of 86%, 43%, and 60%, and a recall of 67%, 60%, and 60% for acne, eczema, and psoriasis. In the federated learning approach, after distributing the dataset among 1000, 1500, 2000, and 2500 clients, the model showed an average accuracy of 81.21%, 86.57%, 91.15%, and 94.15%. The CNN-based skin disease classification merged with the federated learning approach is a breathtaking concept to classify human skin diseases while ensuring data security. Index Terms— Benchmark Algorithms, Convolution Neu- ral Network, Federated Learning Framework, IoMT Security, Medical Imaging, Skin Disease Classification. I. I NTRODUCTION S KIN disease is a terrific and common illness worldwide. There are multiple agents that determine the influence of skin diseases, from environmental factors to genetic susceptibility. Multiple social factors such as poverty, affluence, inequality, education, and access to Manuscript received XXXX, 2021; revised XXXX, 2022; accepted XXXX, 2022. Date of publication XXXX, 2022; date of current version XXXX, 2022. “This work was supported by Researchers Supporting Project number (RSP-2021/100), King Saud University, Riyadh, Saudi Arabia. This work was supported in part by funding from the Natural Sciences and Engineering Research Council of Canada (NSERC).” M. N. Hossen is with the Department of Information and Communica- tion Technology, Mawlana Bhashani Science and Technology University, Tangail-1902, Bangladesh. (e-mail: nazmul.ict92@gmail.com). Vijayakumari P. is with Department of Applied Electronics, Institute of ECE, Saveetha school of Engineering, SIMATS, Chennai, Tamilnadu 602105,India.(e-mail: vijayakumarip. sse@saveetha.com). D. Koundal is with the Department of Systemics, School of Computer Science, University of Petroleum and energy Studies, Dehradun, India. (e-mail: dkoundal@ddn.upes.ac.in). K. Ahmed is with the Department of Electrical and Computer Engi- neering, University of Saskatchewan, 57 Campus Drive, Saskatoon, SK S7N 5A9, Canada and Group of Bio-photomatiχ, Dept. of ICT, MBSTU, Tangail-1902, Bangladesh. (e-mail: k.ahmed@usask.ca). F. M. Bui is with the Department of Electrical and Computer Engi- neering, University of Saskatchewan, 57 Campus Drive, Saskatoon, SK S7N 5A9, Canada. (francis.bui@usask.ca) S. M. Ibrahim is with the Department of Biochemistry, College of Science, King Saud University, P.O. Box: 2455, Riyadh 11451, Saudi Arabia. (e-mail: syakout@ksu.edu.sa). Corresponding Author: K. Ahmed, Dept of ECE, USASK, Saskatoon, Canada. (e-mail: k.ahmed.bd@ieee.org; kawsar.ict@mbstu.ac.bd) Digital Object Identifier 10.1109/JBHI.XXXX health care are all also responsible. According to the Global Burden of Disease (GBD) Study 2010 [1], in the cause of nonfatal diseases in the world as a burden, skin disease was fourth on the list of most common diseases. Skin diseases cause a wide range of problems in both low and high-income countries, including psychological and sociological issues [2]. Its psychological impact is damning. A person with skin disease has anxiety, depression, anger, social isolation, and low self-esteem [3]–[5]. There is an anticipation that skin disease is remediable and medicable if it is detected before a prolonged period. However, dermatologists have a difficult time detecting skin diseases because many of them have the same appearance in terms of color and anatomy [6]. But machine learning has enabled a staggering alteration in medical imaging, especially in disease detection. With the improvement of the processing power of computers and the immeasurable amount of data availability [7], machine learning models have shown human-level activities in medical science. For example, CNN has accelerated progress in medical image processing (e.g., CT scan, MRI) [8]. Clinical images are inadequate for research due to different resolutions, complex contexts, and privacy concerns, particularly with sensitive body part images. Besides, the image of the skin disease dataset is not clearly labeled with information. Moreover, the number of accessible datasets with labeled information is very low. So, research on skin images is troublesome for all the aforementioned conditions. But, there is a problem with machine learning. In it, all the data is gathered in one location, usually a data center, which might possibly breach user privacy and data confidentiality laws. An emerging concept called federated learning will address both of these problems. The data is disseminated across the clients in federated learning, and then a prototype of the central model is delivered to the clients. The transmitting model is trained with client data at each client site, and each client model transmits the update weights or gradient to the central model after a period of time. Finally, the central model is updated using the federated averaging process, and the revised model prototype is provided to clients. II. BACKGROUND STUDY Abundant research articles have been published for skin disease detection and classification. Among them, many researchers have applied deep learning algorithms for skin disease classification [3]. For instance, Esteva et al. [9] showed good accuracy in classifying skin tumors using the V3 inception architecture in 2017. Two dermatological tests showed an accuracy of 55.0% and 53.3% for nine classes of tumors, respectively. The model showed an average accuracy of 55.4%. In the same year, Codella et al. studied a nonlinear support vector machine (SVM) algorithm for detecting melanoma, using 70% of the data for training and 30% of the data for testing [11]. The studies achieved an average accuracy of 76%. In 2018, Zhang et al. [10] also proposed the V3 inception architecture on dermoscopic images using the same network for classifying four common skin diseases such as psoriasis, SK, BCC, and melanocytic nevus. The outcomes achieved an accuracy of 87.25%. The accuracy of the aforementioned studies is never greater than 90%. Shanthi Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on February 10,2022 at 07:24:26 UTC from IEEE Xplore. Restrictions apply.