Figure 1
From: AI-based localization and classification of skin disease with erythema
source implementation of the U-Net (v0.1.2) is available at: https://github.com/qubvel/segmentation_models.pytorch.
Schematic flow. From left to right, the original is first decomposed into hemoglobin and melanin images. All three images are input to the U-Net which outputs a black-and-white mask image. This mask image is used to draw contours each cluster. A convex hull algorithm is applied to crop each cluster. Each cluster is input to the EfficientNet, which generates a prediction alongside the confidence rate. An open-
