Domain Shifts in Dermoscopic Datasets

Abstract

The objective of this work is to estimate domain shifts in dermoscopic datasets from International Skin Imaging Collaboration (ISIC) archive. We analyzed the datasets from different clinics that comprise of various image acquisition systems and lighting conditions. Also, these images are obtained from a wide range of patients with various biological factors like skin color, age and gender. Based on the domain shifts calculated, datasets are split into in and out of domain regimes. Our focus is to quantify domain shifts in dermoscopic datasets and in-turn evaluate their influence on the performance of unsupervised domain adaptation methods. I will present the comparative analysis on how the domain adaptation methods are performing on the domain shifted datasets.