Forum for Nordic Dermato-Venereology No 3, 2019 Telemedicine | Page 16

Carsten Sauer Mikkelsen, Kristian Bakke Arvesen, Peter Bjerring and Luit Penninga – Artificial Intelligence in Dermatology
consistent and rather objective technique. They can assist
dermatologists in different steps of analysis, such as detection
of the lesion boundary, quantification of diagnostic features,
classification into different lesions types (tumour staging) and
visualization. Many studies have been published; a systematic
review has identified 182 publications on the topic between
1985 and 2011 (5, 10, 13). Two main types of computer-as-
sisted dermoscopy exist: computer-assisted dermoscopy based
on dermoscopic images (CAD-derm) and computer-assisted
dermoscopy based on spectroscopy (CAD-spect), which is
predominantly multi-spectral-imaging (5, 10, 13). A Cochrane
review on the use of computer-assisted dermoscopy included
42 studies with a total of 15,938 lesions (10). Twenty-four
studies applied CAD-derm and 18 studies used CAD-spect.
The Cochrane review found that, for a group of 1,000 skin
lesions, of which 200 (20%) are given the final diagnosis of
melanoma, 386 people will have a CAD-derm result suggesting
that a melanoma is present and, of these, 206 (53%) will have
a false-positive result (10). Twenty (3%) of 614 people with a
negative CAD-derm result have a melanoma (false-negative
result). Dermoscopy or CAD-derm were found to be equal in
their ability to detect or rule out melanoma (10).
For a group of 1,000 people, of whom 200 (20%) are given the
diagnosis of melanoma, 637 will have a positive CAD-spect
result, suggesting that a melanoma is present. Of these 451
(71%) will have a false-positive result. Fourteen (4%) of 363
people with a positive CAD-spect result will have a melanoma
(false-negative result) (10). CAD-spect detects more melano-
mas, but possibly produces more false-positive results (an
increase in unnecessary surgery, referrals and worries). The
review concludes that both CAD types demonstrate high sen-
sitivity, and could be used as a back-up for specialist diagnosis
to assist in the diagnosis of melanomas. The Cochrane review
concludes that the available data are too limited to make a
judgement on which method of CAD to use (10).
Another review also evaluated the diagnostic accuracy of
dermoscopy and digital dermoscopy for the diagnosis of
melanoma (13). The authors retrieved 765 articles, and 30
studies were included in their meta-analysis. The meta-ana­
lysis showed that sensitivity for CAD-based dermoscopy was
slightly higher than for dermoscopy (91% vs. 88%; p = 0.076),
while specificity for dermoscopy was significantly better
than CAD-based dermoscopy (86% vs. 79%; p < 0.001). The
diagnostic odds ratio for dermoscopy (51.5) and CAD-based
dermoscopy (57.5) were not significantly different (p = 0.783).
The author concluded that both tests are equal for diagnosis
of melanoma (13).
Based on the literature review, it appears that CAD-based
dermoscopy has become a valuable diagnostic tool, although
100
T heme I ssue : T eledermatology
it needs further development and evaluation. It can currently
be used as a supplement or back-up to regular dermoscopy.
Further research is needed to establish the exact role of CAD-
based dermoscopy.
C onclusion
Applications of AI in the field of dermatology are increasing-
ly used, both by patients and health professionals. Mobile
phone apps with image analysis for patients need further
improvement before they can be considered safe, and not
miss diagnoses of melanoma. CAD-based dermoscopy appears
to have evolved into a valuable tool supplementing regular
dermoscopy.
The majority of current AI applications involve melanoma,
although more AI applications for the diagnosis of dermato-
logical diseases, including other types of skin cancer and other
dermatological diseases, are anticipated.
Ultimately, while reflecting on the potential of AI to improve
the accuracy and efficiency of dermatological diagnosis, we
must always keep in mind the holistic nature of clinical der-
matology with its focus on the whole patient, and on effective,
thorough and compassionate doctor–patient communication.
R eferences
1. Lim BCW, Flaherty G. Artificial intelligence in dermatology: are
we there yet? Br J Dermatol 2019. Available from: https://doi.
org/10.1111/bjd.17899.
2. Global Health Workforce Alliance and World Health Organization.
A universal truth: no health without a workforce. Third global
forum on human resources for health report. Geneva: WHO; 2013.
3. Schofield JK, Grindlay D, William HC. Skin conditions in the UK: a
health needs assessment. 2009. Available from: www.nottingham.
ac.uk/scs/divisions/evidencebased dermatology/news/dermatolo-
gyhealthcareneeds assessmentreport.aspx.
4. Proprietary Association of Great Britain and Readers Digest. A
picture of health: a survey of the nation’s approach to everyday
health and wellbeing. London: Proprietary Association of Great
Britain, 2005.
5. Ali A-R A, Deserno TM. Systematic review of automated mel-
anoma detection in dermatoscopic images and its ground
truth data. Available from: https://pdfs.semanticscholar.
org/e84e/dcb0d71584ffba563088b757379859ebf34e.pdf?_
ga=2.75794514.636114537.1557822888-2117381039.1538020987.
6. Kassianos AP, Emery JD, Murchie P, Walther FM. Smartphone
applications for melanoma detection by community, patient and
generalist clinician users: a review. Br J Derm 2015; 172: 1507–1518.
7. Chuchu N, Takwoingi Y, Dinnes J, Matin RN, Bassett O, Moreau JF,
et al. Smartphone applications for triaging adults with skin lesions
that are suspicious for melanoma. Cochrane Database Syst Rev
2018 Dec 4; 12: CD013192.
8. Dinnes J, Deeks JJ, Grainge MJ, Chuchu N, Ferrante di Ruffano L,
et al. Visual inspection for diagnosing cutaneous melanoma in
adults. Cochrane Database Syst Rev 2018 Dec 4; 12: CD013194.
Forum for Nord Derm Ven 2019, Vol. 24, No. 3