The incidence of skin cancer worldwide has grown at an alarming rate since the mid-1990s. The World Health Organization estimates that the incidence triples every decade. It is believed that in the last four years it has increased by 38% in Spain, according to data from the Spanish Academy of Dermatology and Venereology. In addition, this trend is expected to continue in the next two decades, reason why awareness-raising and prevention campaigns are so necessary.

There are different types of skin cancer. The most well-known for its high malignancy is melanoma, although it has a lower incidence than other types such as squamous cell carcinoma and basal-cell carcinoma (more frequent type). In any case, prevention is based on controlled exposure to solar radiation (especially ultraviolet radiation) and monitoring and control of existing moles and suspicious lesions.

At present, it is estimated that 20% of primary healthcare consultations in Spain refer to dermatology topics. These professionals often hesitate while analysing the lesions and refer the cases to the specialist dermatologist. Due to the increase in consultations of this type, tele-dermatology services are being created in order to speed response times and avoid saturation of the health system.

One of the lesions that cause more confusion during its diagnosis is keratosis, mainly actinic and seborrheic. Actinic keratosis is an over-elevated, rough and poorly delimited lesion that appears on parts of the skin exposed to the sun for a long time or previously sun burned. It is a recurrent lesion that may be considered pre-malignant since it can degenerate into squamous cell carcinoma (estimated in 20% of cases). On the other hand, seborrheic keratosis is a dark appearance lesion with a rough texture that shows a growth and bulging of the skin and can be located anywhere in the body. It is a benign lesion, but because of its unsightly appearance and some of the symptoms, it can be confused with lesions such as malignant pigmented melanoma.

The correct diagnosis of these and other skin lesions is a determining factor in their treatment, especially to discard their potential malignancy. These types of keratosis, in addition of being confused with the above mentioned malignant lesions, may also be similar to benign ones such as: solar lentigo, nevus (moles), verruca vulgaris, dermatofibroma, non-psoriatic eczema or psoriasis

Dermatology diagnosis support tool developed by Tecnalia

Tecnalia’s Computer Vision team is working on the development of a dermatology diagnostic support tool which will facilitate decision-making. In-house and advanced algorithms of image processing based on the paradigm of ‘deep learning’ [1] have been developed. Convolutional networks are used to automatically recognize patterns in large databases to solve problems such as object detection and recognition, segmentation, classification, etc., having achieved unprecedented results with respect to classical machine learning methods. Recently, Stanford University has published a work in Nature [2]where using this type of networks achieve a level of diagnosis comparable to a dermatologist. However, this study focuses only on solving very specific problems, such as the differentiation between melanoma and benign nevi. On the contrary, in the case of Tecnalia’s tool, a higher spectrum of diseases is being considered, as the case of keratosis and the differential diagnoses mentioned above.

This tool follows one of the main objectives of the European Union and health systems within the Horizon 2020 plan: create new diagnostic tools to contribute to the sustainability of the health system and adapt to the growing needs of the population. A tool like this will facilitate the on-site diagnosis of lesions and will reduce the number of specialist referrals. This will not only reduce the health costs associated during the process, but will also result in a minor inconvenience to the patient, with all that this may involve. In addition, the early diagnosis of malignant lesions will allow applying the most appropriate treatment in advance and improve the prognosis of the patients.

[1] Deep learning: Y. LeCun, Y. Bengio, and G. Hinton, “Deep learning,” Nature, vol. 521, no. 7553, pp.436–444, May 2015.

[2] A. Esteva et al., “Dermatologist-level classification of skin cancer with deep neural networks,” Nature, vol. advance on, Jan. 2017.