atopic dermatitis evolution with artificial intelligence

ASCORAD: Automatic grading Atopic Dermatitis with Deep Learning. Cutaneous artificial intelligence for Eczema

Keywords: Atopic dermatitis, SCORAD, Deep learning, Automatic severity assessment, CADx system

ABSTRACT

Atopic dermatitis is a chronic, itchy skin condition that affects 15–20% of children but may occur at any age. It is estimated that 16.5 million U.S. adults (7.3%) have AD that initially began at >2 years of age, with nearly 40% affected by moderate or severe disease. In addition, a meta-analysis including over 110,000 subjects found that 20% of children with atopic dermatitis still had persistent disease 8 years later.

¿Do you want to see the Automatic SCORAD algorithms in action?

Therefore, a quantitative measurement that could track the evolution of Atopic dermatitis severity could be extremely useful on assessing therapeutic efficacy.

Currently, SCORAD (SCOring Atopic Dermatitis) is the most frequently used measurement in clinical practice. However, SCORAD has the following disadvantages:

  • more_timeManually filling the SCORAD is very time consumingUsually takes about 7–10 minutes per patient which poses a heavy burden on dermatologists; and
  • compare_arrowsInconsistencyDue to the complexity of SCORAD calculation, dermatologists can give different scores for the same case

In order to solve these problems, we propose to apply state-of-the-art neural networks to estimate atopic dermatitis severity based on skin lesion images. Different from typical deep learning frameworks for image processing, our network SCORADnet is able to estimate surface and all visual signs at once, improving the efficiency at both tasks simultaneously.

scoring atopic dermatitis
Screen capture of the algorithm output for atopic dermatitis severity estimation

The experimental results show that SCORADnet can achieve the mean absolute percentage error of 13%, outperforming baseline methods and below the inter observer variability of 20%.

SCORADnet is able to estimate surface and all visual signs at once, improving the efficiency at both tasks simultaneously

Overall, SCORADnet helps practising evidence-based dermatology, relieves dermatologists from the tedious SCORAD calculation, enables patients to track Atopic dermatitis severity in a much user-friendly and objective way and enables a more precise evaluation of new treatments.

¿Do you want to see the Automatic SCORAD algorithms in action?

Automatic scoring for atopic dermatitis
Figure 1. The visual signs that compose the SCORAD with four intensity levels: none, mild, moderate and severe. We trained six EfficientNet-B0 networks, one for each visual sign severity assessment

Get access now

Fill in the form to get access to the tool.

Leave a comment