Dermatology has evolved over the years, but the common thread remains visual literacy. Visual literacy is the learning, understanding, and interpretation of images by the observer.  Visual literacy is key to the observation of art and the effectiveness of artificial intelligence systems in dermatology.
Dermatology is a visual specialty that relies on pattern recognition to make a diagnosis. These visual skills may be enhanced by understanding and appreciating art. As part of their training, dermatology residents in the US, took an art appreciation course at the museum of arts in Boston.  Students were given both clinical and art images to interpret before and after the course and results showed an improvement in visual diagnostic skills. Similar pilot studies have been trialled in the U.K with a perceived improvement in visual skills by participants. [3,4] Reflection also improved, which in medicine is key when evaluating a new patient.
As dermatologists, we look at a patient, pause and consider the history and investigations, and look again to make a differential diagnosis. Experienced art observers often do not accept their first impression of an image, but examine the work in context of the artist’s style, the name, and backstory of a piece of art, before making their final interpretation. For example, Edward Hoppers’ famous “Nighthawks” painting is a vision of New York in the 1940s.  On first impressions, the painting may be interpreted as moody and uninteresting, but in context of the title, knowledge of Hoppers’ style, closer inspection of individual figures and the dramatic contrast with light and shade; a more powerful, intriguing and possibly ominous piece can be seen. However, the visual literacy of a painting can be subjective, as can the visual literacy by a dermatologist and a patient with skin disease.
The use of artificial intelligence in medicine is potentially ground-breaking and is particularly relevant to those specialties that rely on pattern recognition to make a diagnosis. Artificial intelligence systems have successfully been trialled in radiology, oncology, ophthalmology, and gastroenterology to aid diagnosis. It is hardly surprising that dermatology, a visual specialty; clinically, dermatoscopically, and histologically, lends itself to this technology.
Artificial intelligence systems are based on artificial neural networks which represent the workings of the normal brain. Huge amounts of labelled data (ground truth) are fed into the neural networks, which learn, extract information from the images, and produce algorithms in order to independently make decisions on classification of new images. This is a form of visual literacy by the artificial intelligence system; they have learnt patterns which distinguish images and are able to make independent diagnoses due to this learned behaviour. However, the precision of an artificial intelligence system is based on the accuracy and amount of data inputted. For example, one could confidently teach an artificial intelligence system to distinguish melanomas from benign naevi, as large datasets of images are achievable, but it will be difficult to teach an artificial intelligence system to diagnose rare skin cancers with only a limited bank of images. Humans can see an image once or twice and remember it forever.
Skin cancer has risen by 20% per year for the last decade in the UK and this demand is overwhelming many dermatology departments.  Detection of skin cancers is an obvious focus for artificial intelligence in dermatology. Research has shown that artificial neural networks can be as accurate as 21 dermatologists in diagnosing benign and malignant skin lesions.  Further research showed that artificial intelligence outperformed dermatologists in diagnosing benign and malignant skin lesions.  These findings could result in a potential solution to the two week wait skin cancer crisis.
However, the question arises, whether artificial intelligence systems could be developed which are able to diagnose the more complex or rare disorders. Takayasu’s Arteritis presenting with fevers, aortic regurgitation, and skin necrosis may require input from multiple specialties and numerous investigations before the diagnosis can be reached.  In addition, images of skin disease in Fitzpatrick darker skin types IV-VI are limited, and unless an Artificial Intelligence system is trained with the appropriate “ground truth,” diagnostic accuracy would be affected. However, there is a growing body of literature in ethnic dermatology, which aims to improve both the clinician’s diagnostic skills as well as Artificial Intelligence systems. 
As clinicians, we have an emotional quest to make the right diagnosis for our patients and make them better. Artificial intelligence systems in dermatology do not have this emotional connection and their diagnoses are based on visual images alone. The dermatologist should not fear replacement by artificial intelligence technology, but view this advance as augmented intelligence, where man and machine assist each other to improve patient care. The universal rapid growth of artificial intelligence is inevitable and exciting, but there needs to be better understanding of both accuracy and accountability before it reaches the clinic.
Visual literacy is integral to the dermatologist. It is a human skill that can be heightened by understanding and appreciating art. Artificial intelligence may attempt to learn it, but is unlikely to be able to replicate the skill completely. The modern dermatologist would benefit from improving their visual literacy, but should also embrace the many benefits that will come from artificial intelligence.
Monika Saha, consultant dermatologist, Lewisham and Greenwich NHS Trust
Mo Saha, creative Strategy Partner of WorkSmiths, a brand and innovation studio London.
Competing interests: none declared.
- Debes JL. The loom of visual literacy- an overview. Audiovisual Instr 1969; 14,8, 25-27
- Zimmerman C, Huang TJ, Buzney EA. Refining the eye: Dermatology and visual literacy. J of Museum Education 2016; 41: 116-122
- Griffin LL, Chiang NYZ, Tomlin H et all. A visual literacy course for dermatology trainees. Br. J Dermatol. 2017;177 (1): 310-311.
- Gardham N, Walsh S. Comment on ‘A visual literacy course for dermatology trainees. Br J Dermatol 2018;178(2):572-573.
- Eedy D. The crisis in dermatology. BMJ 2015;350:h2765.
- Esteva A, Kuprel B, Novoa RA et al. Dermatologist-level classification of skin cancer with deep neural networks. Nature 2017; 542 (7639): 115-118.
- Brinker TJ, Hekler A, Enk AH 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.
- Estfan Y, D’Cruz DP, Patel S et al. An unusual and potentially fatal cause of scalp crusting. Clin Exp Dermatol 2017; 42(4): 441-443.