Automatic Detection of Skin Surface Structure Using Deep Learning for the Impression Mold Technique

This paper presents a discriminator that automatically discriminates skin ridges and folds by performing image analysis using a discriminator that uses deep learning. We use U-Net as the discriminator and a data set including images of the thighs, forearm, chest, forehead, and other parts of healthy controls and atopic dermatitis. We evaluate the discriminator using the mDice coefficient, which is the average of the Dice coefficient of the skin ridges and the Dice coefficient of the skin folds of each part. As a result, for 96 images of a test data, the average mDice coefficients for healthy controls and atopic dermatitis are 0.92 and 0.96, respectively.