Applications of Artificial Intelligence to Dermatology

The AI-integrated Health Service of the Future?

There are attempts to deploy 'AI' technologies within the healthcare space within two main scenarios: direct to consumer or public, and as a decision aid for clinicians. The direct-to-consumer model already exists in some fashion; there are smartphone apps such as SkinVision, which enable individuals to assess and track their skin lesions. However, currently such apps do not make accountable diagnoses and usually explicitly state in their terms and conditions that they do not provide a diagnostic service, and do not intend to replace or substitute visits to healthcare providers. At present, it is not yet clear what the benefits and risks of such a tool are in terms of how frequently it provides false reassurance, and how frequently it recommends referral when this is not needed. Although health data democratization has benefits from the perspective of patient autonomy, it may be that this does not translate to better health outcomes and might instead lead to unnecessary concern and investigations. Moreover, fundamentally, healthcare is currently structured in such a way that responsibility and liability are carried by the provider and not the patient, and as such these apps do not have a clear-cut position in healthcare infrastructure.

The current social and legal framework of healthcare is better primed for incorporating AI as a decision aid for clinicians, particularly in enhancing decision making by nonspecialists (Figure 5). This could potentially be of great use in dermatology services due to the ever-growing burden of skin cancer. In the UK, there is a long-standing shortfall of consultant dermatologists, and current workforce planning is insufficient to address this. The volume of skin cancers has a knock-on effect on patients with chronic inflammatory skin diseases, essentially reducing their access to dermatologists.

(Enlarge Image)

Figure 5.

Schematic showing hypothetical use of a machine learning algorithm to help nonexpert clinicians risk-stratify lesions to make clinical decisions. Clinicians routinely weigh up both the benefits and limitations of common diagnostic aids such as prostate-specific antigen or D-dimers. Currently, there are very few useful dermatological diagnostic decision aids available to nonexpert clinicians, as the diagnostic process is dominated by image recognition. Convolutional neural network could represent a new class of decision aid that could help nonexpert clinicians triage appropriately and narrow down their differential diagnosis.

Dermatologists are also aware that generally, a high proportion of referrals to dermatology with suspected skin cancer on the urgent '2-week wait' pathway do not require further investigation and are actually immediately discharged. Many of the lesions falling into this category are easily recognized by dermatologists, but are not easily recognized by nonspecialists. One could hypothesize that CNN-based applications can aid a general practitioner service in triaging skin lesions more effectively, and ensure that patients are managed by the appropriate clinical services. Having a clinical user also mitigates many of the risks and limitations inherent to CNN-based technologies, improving both the safety profile and the patient experience.

The recently published Topol Review on 'Preparing the healthcare workforce to deliver the digital future' states that 'to reap the benefits, the NHS must focus on building a digitally ready workforce that is fully engaged and has the skills and confidence to adopt and adapt new technologies in practice and in context'. It also concludes that 'the adoption of technology should be used to give healthcare staff more time to care and interact directly with patients'.^[29] In the context of dermatology, this very much holds true. Technology adoption could improve clinical pathways, and enable our neediest patients to access dermatology services more efficiently. It is unlikely that they will threaten our profession; in reality they represent an opportunity for personal learning, service improvement and leadership that could be transformative for our future healthcare system.

Funding sources
X.D-H. is the recipient of an Accelerator Award from Cancer Research UK. F.M.W. gratefully acknowledges financial support from the UK Medical Research Council (MR/PO18823/1), the Biotechnology and Biological Sciences Research Council (BB/M007219/1) and the Wellcome Trust (206439/Z/17/Z). This work was supported by the Francis Crick Institute, which receives its core funding from Cancer Research UK (FC010110), the UK Medical Research Council (FC010110) and the Wellcome Trust (FC010110). N.M.L. is a Winton Group Leader in recognition of the Winton Charitable Foundation's support towards the establishment of the Francis Crick Institute. N.M.L. is additionally funded by a Wellcome Trust Joint Investigator Award (103760/Z/14/Z), the MRC eMedLab Medical Bioinformatics Infrastructure Award (MR/L016311/1) and core funding from the Okinawa Institute of Science & Technology Graduate University. M.D.L. gratefully acknowledges financial support from the Wellcome Trust (211276/E/18/Z).

Table 1. Essential terminology in the field of machine learning and artificial intelligence
Artificial intelligence (AI)	The ability of machines, such as computers, to simulate human intelligence
Machine learning	Algorithms and statistical models that are programmed to learn from data, therefore recognizing and inferring patterns within them. This enables computers to perform specific tasks without explicit instructions from a human operator
Supervised learning	Refers to machine learning tasks whereby the goal is to identify a function that best maps a set of inputs (e.g. image) to their correct output (label). This is based learning or training on prematched pairs. This is in contrast to unsupervised learning, where novel patterns such as groups or 'clusters' are identified in data without influence from prior knowledge or labelling
Overfitting	A common problem in machine learning where the model has high accuracy when tested on data from the same source as its training data, but its performance does not generalize to novel sources of data
Neural network	A form of supervised machine learning inspired by biology whereby data pass through a series of interconnected neurons, which are individually weighted to make predictions. During training, the data pass through the network in an iterative manner and the weightings are continually adjusted to optimize its ability to match label to data
Deep learning	Refers to a neural network with multiple layers of 'neurons' that have adjustable weights (mathematical functions)
Convolutional neural network	Refers to a type of neural network whereby the layers apply filters for specific features to areas within an image