Fig. 2: Relationships between TEWL and features of skin images extracted using the kNN density estimator.

a Typical examples of skin images. Cases A and B show relatively regular skin texture and low TEWL, and cases E and F show relatively rough skin surfaces and high TEWL. The written consent was obtained for publication of the photographs. b The persistence diagrams of cases A and E with marginal distributions. It is remarkable that the mid-life of case A has a sharper peak with a higher value range than that of case E. c The t-value of each variable is calculated using simple linear regression predicting TEWL. As explanatory variables, we used the mean and standard deviation of mid-life and life-time for 0-dim and 1-dim topological features, the number of all connected components and holes, age, and sex. Variables with a false discovery rate (FDR) larger than 0.01 are colored red (if its t-value is positive) or blue (if its t-value is negative). d, e The results of simple linear regression predicting TEWL from the mean and the standard deviation of 1-dim mid-life with the coefficient of determination (R²) and p-value. Since sex is the most important variable in determining TEWL other than the topological features of the images, each point is colored red or blue according to the subject’s sex. The points representing cases A–F are also labeled. The shaded areas represent the 95% confidence intervals of the regression. f The mean and the standard deviation of 1-dim mid-life are linearly related, and a clear observed TEWL trend change can be seen along the regression line. The shaded area represents the 95% confidence interval of the regression.