Fig. 2: Feature representation of the 22 diseases included in our study. | npj Digital Medicine

Fig. 2: Feature representation of the 22 diseases included in our study.

From: Optimizing skin disease diagnosis: harnessing online community data with contrastive learning and clustering techniques

Fig. 2

a t-SNE plots of the filtered training set with 50 validation images per category colored by disease categories, demonstrating the effectiveness of our approach in capturing the distinct features of each disease. For comparison, a subset of images of the same number is randomly selected from the coarse labeled training set. b Average sensitivity and specificity for each disease of 3 trials. Specificity of each category are all over 0.94 but sensitivity varied considerably. Generally, sensitivity would be higher if the cluster was relatively far from the center of other diseases with fewer surrounding clusters. c Expression levels of some host features. The levels of these characteristics are associated with the types of skin diseases and affected areas of the skin.

Back to article page