How can artificial intelligence be used to detect melanoma?

How can artificial intelligence be used to detect melanoma?

How can artificial intelligence be used to detect melanoma?

Of all skin cancers, melanoma is certainly the most aggressive and dangerous lesion. It is not the most common skin cancer, but because of its tumorous nature, it is the one to be most wary of, as it can easily spread and cause secondary lesions to appear in the rest of the body, lesions that can ultimately lead to the patient’s death.

Melanoma is often associated with skin cancer, but in some rare cases it can also affect non-cutaneous areas such as mucous surfaces (e.g., vaginal or oral).

There are several risk factors that can lead to the development of melanoma: for example, age (although half of all melanomas are seen before the age of 50), the presence of moles (although most melanomas occur outside of a mole), skin and hair type (fair skin, blue eyes, red hair), exposure to the sun, use of tanning salons, sunburn, etc.

In medicine, prevention is the key word, and it is important, especially if there are several risk factors, to remain vigilant and to examine one’s own skin and consult a dermatologist.

In these complicated times related to the COVID-19 viral pandemic, it can be more difficult to consult your medical specialist. It is therefore important to be able to do at least some of the work yourself and to observe any suspicious appearance on your skin.

To determine the suspicious nature of a skin lesion or mole, skin specialists use a set of elements that will help determine whether the lesion is benign or malignant.

The elements used concern the characteristics of the skin lesion: symmetrical or rather asymmetrical character where one of the halves does not correspond to the other half, the aspect of the edges (regular or irregular), the colour of the skin ‘spot’ (single colour or several shades in the lesion), the size of the lesion and its evolution over time.

The dermatologist is the doctor of reference for analyzing this type of situation and should always be consulted in case of any hesitation. For several years, more and more scientific publications and articles in the general press have been highlighting computer tools available on smartphones or in the context of research protocols. These tools can be used to determine the nature of the skin lesion directly or via remote access to a doctor. Some of these tools use artificial intelligence with deep learning and convolutional neural network (CNN) programs.

These are in fact very complex computer programs created to recognize certain particular characteristics of an image (in this case the skin lesion). These programs are based on the recognition of a ‘pattern’ or typical aspect and must be ‘educated’ by means of thousands of different images. This is the same type of algorithm that will allow Facebook to recognize nude images and ban them directly, without the intervention of a human being.

To train these artificial intelligence tools, the researchers will use thousands of images of skin lesions of all types, informing the program that these images correspond either to non-suspicious lesions or to tumoral lesions. After several tens of thousands of images have been ‘ingested’ by the computer software, it will be able to make the final diagnosis itself because it will have learned. It will have learned to make mistakes and to correct those by itself to arrive at the final correct diagnosis. This is what is planned in theory, and recently published medical studies show that these programs are sometimes (or even often) better than the analysis carried out by the specialist doctor. However, not all these programs have yet been used and validated in the public domain and some are still in the research stage. Medical specialists recommend that they be used with caution, for monitoring, support, and recommendation purposes. The last word (for now) still belongs to the human being and to the dermatologist and his or her trained eye.

On the other hand, there are other applications available that will only – without the intervention of artificial intelligence and without providing diagnostic support – allow the evolution of a skin spot or a mole to be monitored. For example, it will suffice for the patient to take a photo of a mole at regular intervals and the software will analyze the evolution and possible modification of this mole (colour, aspect, size, etc.).

Here are some applications available on smartphones that can be used to help diagnose melanoma and potentially suspicious moles:

iSkin: this is an application that allows you to take photos of skin lesions for monitoring over time. This application also allows you to send photos to a dermatologist and carry out a medical consultation remotely.

Skinvision: this is an artificial intelligence algorithm developed with the support of several dermatologists. This paying application does not replace a real medical consultation but allows to obtain very quickly a clue on the suspicious nature or not of a skin lesion.

SkinApp: this is another artificial intelligence algorithm developed in France that helps doctors make a final diagnosis. According to the designer of this tool, the company ANAPIX: “SkinApp was designed as a tool for dermatologists and not as a substitute for their work and expertise. The objective is to assist them in the detection of skin cancers and other skin pathologies by providing them with innovative and easy-to-use tools”.

UMSkinCheck: UMSkinCheck is a free application available on a smartphone that allows users to examine their own skin and follow the evolution of suspicious lesions over time. Application developed by the University of Michigan Medicine in the United States.


What is AI? Applications of artificial intelligence to dermatology. X. Du-Harpur, F.M. Watt, N.M. Luscombe, M.D. Lynch. British Journal of Dermatology (2020)183, pp423–430.

Dermatologist-level classification of skin cancer with deep neural networks. Andre Esteva, Brett Kuprel et ali. Nature. 542, pages 115–118 (2017).

Deep neural networks are superior to dermatologists in melanoma image classification.

Brinker TJ, Hekler A et ali. Eur J Cancer. 2019 Sep;119:11-17

Integrating Patient Data Into Skin Cancer Classification Using Convolutional Neural Networks: Systematic Review. Julia Höhn, Achim Hekler et ali. J Med Internet Res. 2021 Jul 22;23(7):e20708


Related Posts