A digital hand in dermatology

Technology has been incorporated in each and every field humans desire to control, marking an ongoing revolution. Nowadays, digital solutions have become the main player in this transformation and artificial intelligence the leading tool to accomplish it. In medicine, the fields that strongly rely on visual or image screening such as dermatology are the first to undergo the translation of these technologies (1). I remember in my previous work experience the moment a new department’s purchase arrived: an automized machine for performing complete urinalysis. It has a microscope within which would photograph the urine samples and assess it, thus performing the sediment analysis that before could only be accomplished by humans. That is an example of machine learning applied for image screening, how? Because that machine was able to recognize the different elements of urine thanks to its intelligent algorithm that allowed it to learn pattern recognition during training. Artificial intelligence (AI) is the machine’s ability to understand its environment, learn, and operate accordingly in an independent manner, thus mimicking human intelligence. However, that level is still far beyond the range of the current methods: Actually, the term AI is commonly used for related techniques such as machine learning (ML), deep learning and neural networks. ML is the ability of a machine to learn from data due to computational algorithms and statistical models, that allow it to recognize and infer patterns (2). These AI technologies have an essential role in digital medicine and create several opportunities for improving dermatology, which goes even beyond image screening.

An AI technology called “deep convolutional neural network” that is able to classify the most common types of skin cancer and differentiate malignant and benign skin lesions was developed. In addition, its average accuracy is equivalent or higher compared to an experienced dermatologist. Therefore, this computational method allows better tracking of skin lesions, thus early detection of cancer (3), which is the main factor in prognosis. AI-based solutions can be applied to other dermatological conditions such as inflammatory skin diseases, ulcers, and allergy. Moreover, besides image classification, those algorithms can perform risk assessment calculations (4). Thus, AI-based applications can be a fantastic tool for the diagnose and assessment of dermatoses, which can contribute to the health service in many ways. Including triage, assistance in clinical decision-making, and direct use from the individuals (2). Tools that use machine learning in dermatology are under study for several applications, such as melanoma screening, wound assessment, risk predictions, automation in histology, and disease classification. Which can be useful for a number of diseases including psoriasis, acne, lichen planus, pityriasis lichenoides and dermatomyositis. An algorithm was able to predict patient’s responses to biologic therapy and to infer the severity in psoriasis, with high sensitivity and specificity. Machine learning together with molecular biology can provide important risk assessments, such as an AI model that was able to diagnose psoriasis vulgaris with an accuracy of 96.4%, based on gene expression evaluation. Another similar study was able to predict the tendency to developing psoriatic arthritis symptoms in cutaneous psoriasis patients. Furthermore, the use of machine learning can have an impact on drug discovery and development. For instance, being applied in predicting skin sensitization substances, which could decrease animal testing (4). Another important aspect is the possibility to involve patients with their own health. One method to accomplish that is the direct use of these technologies by the patients. For instance, intelligent algorithms capable of assessing and tracking skin lesions were developed for smartphones (2). A group developed a deep-learning algorithm for assisting the diagnose of onychomycosis which can be potentially used for mobile devices (5).

The potential of AI-based technology in dermatology is limitless and inevitable. However, it does not mean that there are no issues or concerns associated with it. The implementation itself is already complicated considering the following matters (1):

· Data sharing and privacy: as mentioned before, the algorithm needs data to be trained but not only that. Data supply has to be continuous in order to maintain the AI technology updated and to allow improvements. However, sharing data among institutions and nations can be complicated for many reasons. In addition, data gathering from individuals is a barrier when privacy is protected by regulations.

· Transparency: algorithm’s transparency is important for interpretation and supervised learning. Supervised learning implies the accuracy of the algorithm’s predictions, while interpretation allows investigation, thus correction of algorithmic bias.

· Data standardization: allows compatibility across multiple platforms, thus a common format for data is important for AI technology’s efficiency; however healthcare data is usually heterogeneous.

· Patient safety: the current regulatory framework does not offer clear quality control standards for algorithms and criteria to evaluate AI technology’s safety and efficacy.

Besides the mentioned complications, the problem of accountability is pressing. Who is responsible for the outcomes of these technologies? In terms of liability, the AI applications do not have a clear position in the healthcare infrastructure yet (2). Furthermore, there is little collaboration from dermatologists in the development of AI technologies (4), which can impair the quality of the tool.

AI-based technologies are powerful tools in medicine, but especially in dermatology. The synergy between health professionals and technology can uplift the field. Although most studies report its applications mainly in the diagnose and managing of the diseases, one can hypothesize that those technologies may reach other fundamental components of health. Digital medicine is also expected to increase the engagement of patients with their health (6). Nevertheless, there is still much work to be done regarding implementation, such as technological and clinical validation studies (4). In addition, future dermatologists should receive special training that enables a better understanding of those technologies. The path dermatology already started to follow is clear and it is digital.

Bibliography

1. He, J. et al. The practical implementation of artificial intelligence technologies in medicine. Nat. Med. 25, 30–36 (2019).

2. Du-Harpur, X., Watt, F. M., Luscombe, N. M. & Lynch, M. D. What is AI? Applications of artificial intelligence to dermatology. Br. J. Dermatol. 1–8 (2020) doi:10.1111/bjd.18880.

3. Esteva, A. et al. Dermatologist-level classification of skin cancer with deep neural networks. Nature 542, 115–118 (2017).

4. Gomolin, A., Netchiporouk, E., Gniadecki, R. & Litvinov, I. V. Artificial Intelligence Applications in Dermatology: Where Do We Stand? Front. Med. 7, 1–7 (2020).

5. Han, S. S. et al. Deep neural networks show an equivalent and often superior performance to dermatologists in onychomycosis diagnosis: Automatic construction of onychomycosis datasets by region-based convolutional deep neural network. PLoS One 13, 1–14 (2018).

6. Medicine in the digital age. Nat. Med. 25, 41591 (2019).