|Year : 2020 | Volume
| Issue : 1 | Page : 1-4
Machine learning for potent dermatology research and practice
Ahmed Al-Imam1, Faris Al-Lami2
1 Department of Anatomy and Cellular Biology, College of Medicine, University of Baghdad, Iraq; CERVO Brain Research Centre, Faculty of Medicine, University of Laval, Canada; IBM-Certified Specialist: SPSS Statistics, IBM, United States
2 Iraq Field Epidemiology Training Program (I-FETP) Residents Adviser; Department of Community and Family Medicine, College of Medicine, University of Baghdad, Iraq
|Date of Submission||15-Sep-2019|
|Date of Acceptance||11-Nov-2019|
|Date of Web Publication||27-Mar-2020|
Department of Anatomy and Cellular Biology, College of Medicine, University of Baghdad; CERVO Brain Research Centre, Faculty of Medicine, University of Laval; IBM-Certified Specialist: SPSS Statistics, IBM, United States
Source of Support: None, Conflict of Interest: None
|How to cite this article:|
Al-Imam A, Al-Lami F. Machine learning for potent dermatology research and practice. J Dermatol Dermatol Surg 2020;24:1-4
| What Is Machine Learning?|| |
Machine learning is a rapidly growing application of artificial intelligence (AI) that analyzes big data using multiple mathematical models, including neural networks, regression analysis, and classification trees. AI attempts to minimize the prediction error of mathematically oriented causality associations. Machine learning is particularly useful for assessing spatiotemporal phenomena, including epidemiological investigations., The infrastructure of big data on which AI operates is the same as those used in classical epidemiological research. Researchers and practitioners can assess data from surveys and internet snapshots, longitudinal and cross-sectional studies, and social networks. Forward-thinking researchers combine classical research methods with machine learning. Dermatology researchers can use machine learning to assess the dermatological diseases at national and international levels.
| Machine Learning in Medicine|| |
Data science and machine learning are advancing rapidly. Machine learning is revolutionizing medical and paramedical research. However, the implementation of AI technologies may be lagging in dermatology research and practice. Machine learning can be used to diagnose a broad array of common and rare skin disorders.,,, Besides, AI can be more accurate than dermatologists are. AI, targeting the lowest possible rates of an error on prediction via gradient descent algorithms, can achieve the highest accuracy when compared to classical methods of statistical inference based on hypothesis testing. Although researchers have implemented the use of real-time analytics and predictive models in various specializations of mathematical and natural sciences, the analog implementation of those techniques is still lagging in describing, assessing, and anticipating phenomena connected to dermatological research and practice. Researchers can use machine learning tools in dermatology for data retrieval, the systematic literature review, and prediction of future trends of dermatological conditions and their corresponding spatiotemporal epidemiological patterns.
| Machine Learning in Dermatology: The Status Quo|| |
During the first half of September 2019, we performed a systematic evidence-based review of the published literature to assess the extent of machine learning-based studies in dermatology research. We searched PubMed (the United States National Library of Medicine), the Cochrane Library (the Cochrane Collaboration), and Embase (Elsevier Database), by deploying a detailed set of MeSH-based keywords and generic terms, in combination with Boolean operators as well as truncations, to retrieve potential papers that implemented AI. We included the keywords related to machine learning, data mining, real-time analytics, and predictive modeling for big data in connection with dermatology research and practice [Table 1]. The review strategy generated a total of 1,832,931 papers, allocated into the United States National Library of Medicine (1,614,840 [88.10%]), Embase (210,654 [11.49%]), and the Cochrane Library of Systematic Reviews (7437 [0.41%]) [Figure 1]. Based on the full combination of keywords, we excluded 55,857 papers allocated to Embase, as the indexed volume included false-positive data signals as a reflection of keywords non-sensitive search results, as well as duplicates across the databases. Subsequently, we used the combinatorial of thematic keywords search of the literature to limit the search volume to 19 papers indexed in PubMed (13) and the Cochrane Library (6) [Table 1]. Following the full-text retrieval, there were only 12 peer-reviewed articles (12 [63.16%]) that were relevant to our research question. These were original articles that deployed machine learning, and all are indexed in PubMed and the Cochrane Library [Table 2]. Twenty-five percent of the studies dealt with psoriasis, while the rest concerned atopic dermatitis, Buruli ulcer, ingrown nails, basal cell carcinoma, Merkel cell carcinoma, arsenic-induced skin lesions, cutaneous infections, pigmented skin lesions, and fungal skin infections. Half of the publications originated from the United States, while the rest were from Turkey, Benin, Italy, Japan, Spain, and France. The research output included cross-sectional studies (3 papers, 25%), retrospective designs (3, 25%), prospective cohorts (2, 17%), cross-over experiments (1, 8%), prospective controlled analytics (1, 8%), and randomized controlled trials (2, 17%). There were neither systematic reviews nor metanalytic studies. All of the studies implemented regression models which represent the archetypal machine learning method. Only one study, from Japan, deployed the use of neural networks in connection with atopic dermatitis in infancy. In 1999, Takahashi et al. developed a predictive model for atopic dermatitis in infancy by implementing neural networks and multiple logistic regression. Two decades later, Brinker et al. published a milestone paper at the European Journal of Cancer, in which they found that deep learning outperformed 136 of 157 dermatologists (86.62%) in dermoscopic image classification task for patients with malignant melanoma.
|Table 1: Keyword-based investigation of the principal databases of literature for medical research|
Click here to view
| Machine Learning for Potent Dermatology Research|| |
High-impact researchers can implement a high-precision and systematic methodology in connection with machine learning and AI [Figure 2] by planning a stepwise approach to (1) Conduct a rigorous systematic review of the existing databases of literature in connection with their primary objective; (2) fine-tune the systematic retrieval and appraisal of the literature by deploying automated, non-human mediated, real-time algorithms; (3) assess the prevalence of dermatology-related phenomena via cross-sectional, longitudinal studies, and clinical trials; (4) compare the results with substantiative data from collateral online resources of interest, including Google Trends and Google Analytics open-source deposits of big data to assess the digital epidemiology and geographic mapping of the phenomena; (5) build an all-inclusive multivariate predictive model to anticipate sporadic dermatologic conditions as well as crisis events, including epidemics; (6) attempt to enhance the predictive power by refining the mathematical basis of the predictive models using reliable data transformers to infer robust statistics while minimizing the computational time-cost function for analyzing exhaustive sets of big data and potentially in real-time;, (7) experiment with a variety of non-Bayesian statistical methods to extrapolate data on the patterns of dermatological diseases using between- and within-subject statistical analyses;, (8) evaluate the predictive tools in connection with their sensitivity and specificity, predictive values (positive and negative), the magnitude of error, and the statistical accuracy;, (9) execute concepts of integral mathematics in receiver operating characteristics (curve analysis) for an assessment of the prognostic precision of the tools; (10) enact the models to foresee an imminent geo-specific crisis and future dermatology-related epidemics; (11) create predisaster protocols for the execution by health-care workers, including diagnosticians and interventionalists; (12) formulate empirical and externally valid guidelines for managing the worst-case scenario in disaster situations through collaboration with public and private health authorities; and (13) disseminate the knowledge by communicating collaborative and evidence-based expertise to global regulatory agencies and health organizations, thus aiming for the universal diffusion of unbiased and applicable information.,,
|Figure 2: Penta-hierarchial stepwise approach for potent dermatology research.†Created with CMapTools [version 6.03.02]|
Click here to view
| References|| |
Bishop C. Pattern Recognition and Machine Learning, 1st ed. New York: Springer-Verlag New York; 2006.
Witten I, Frank E, Hall M. Data Mining: Practical Machine Learning Tools and Techniques, 3rd ed. San Francisco: Morgan Kaufmann; 2011.
Lin SY, Mahoney MR, Sinsky CA. Ten ways artificial intelligence will transform primary care. J Gen Intern Med 2019;34:1626-30.
Wiens J, Shenoy ES. Machine learning for healthcare: On the verge of a major shift in healthcare epidemiology. Clin Infect Dis 2018;66:149-53.
Jackson PC. Introduction to Artificial Intelligence, 3rd
ed. New York: Dover Publications; 2019.
Al-Imam A. A systematic literature review on delusional parasitosis. J Dermatol Dermatol Surg 2016;20:5-14.
Kridin K, Shihade W, Bergman R. Mortality in patients with bullous pemphigoid: A retrospective cohort study, systematic review and meta-analysis. Acta Derm Venereol 2019;99:72-7.
Al-Imam A, Al-Shalchi A. Ekbom's delusional parasitosis: A systematic review. Egypt J Dermatol Venerol 2019;39:5-13. [Full text]
Brinker TJ, Hekler A, Enk AH, Klode J, Hauschild A, Berking C, et al.
Deep learning outperformed 136 of 157 dermatologists in a head-to-head dermoscopic melanoma image classification task. Eur J Cancer 2019;113:47-54.
Al-Imam A, Khalid U, Al-Hadithi N, Kaouche D. Real-time inferential analytics based on online databases of trends: A breakthrough within the discipline of digital epidemiology of dentistry and oral-maxillofacial surgery. Mod Appl Sci 2019;13:81-94.
Takahashi K, Hayasawa H, Tomita M. A predictive model for affect of atopic dermatitis in infancy by neural network and multiple logistic regressio. Arerugi 1999;48:1222-9.
Takeda M, Ina H, Kobayashi S. Fourier-transform method of fringe-pattern analysis for computer-based topography and interferometry. J Opt Soc Am 1982;72:156-60.
Razali NM, Wah YB. Power comparisons of shapiro-wilk, kolmogorov-smirnov, lilliefors and anderson-darling tests. J Stat Model Anal 2011;2:21-33.
Cohen J. Statistical power analysis. Curr Dir Psychol Sci 1992;1:98-101.
Gelman A. Objections to bayesian statistics. Bayesian Anal 2008;3:445-9.
Kraemer HC, Blasey C. How Many Subjects?: Statistical Power Analysis in Research, 2nd
ed. Los Angeles: SAGE Publications; 2015.
Al-Imam A. Monitoring and Analysis of Novel Psychoactive Substances in Trends Databases, Surface Web and the Deep Web, with Special Interest and Geo-Mapping of the Middle East. Info: Eu-Repo/Semantics/masterthesis [Dissertation on the Internet]. United Kingdom: University of Hertfordshire; 2017. [DOI: 10.13140/RG.2.2.27636.24961].
Al-Imam, A. Inferential analysis of big data in real-time: One giant leap for spatiotemporal digital epidemiology in dentistry. Odontostomatol Res Anat Learn Implantol 2019;12:1-14.
Al-Imam A. Novel Psychoactive Substances Research: On the Necessity of Real-time Analytics and Predictive Modelling. Res Adv Psychiatry. 2020;7.[In press].
[Figure 1], [Figure 2]
[Table 1], [Table 2]