Infantile Hemangiomas Segmentation using U-Net and SegNet Architectures Models

Infantile hemangiomas are benign tumors that can appear at birth or in the first months of life. As there is no robust protocol in medical practice to determine the evolution of the lesion, an automatic segmentation method can assist physicians to monitor and assess the hemangioma status. Current study proposes two CNN based segmentation approaches using U-Net and SegNet architectures applied on a small proprietary database consisting in 220 images from 35 patients. We have calculated six metrics to evaluate and compare the performance of both networks: accuracy, negative predicted values, positive predicted values, sensitivity, specificity rates and F1-score.