Identifying chaos in dermoscopy images with a deep convolutional network
Title: Identifying chaos in dermoscopy images with a deep convolutional network
SNIC Project: LiU-compute-2020-17
Project Type: LiU Compute
Principal Investigator: George Osipov <george.osipov@liu.se>
Affiliation: Linköpings universitet
Duration: 2020-06-06 – 2020-11-01
Classification: 10201
Keywords:

Abstract

Our project is looking into an multiple step procedure that dermatologist use for detecting potential malignant skin lesions. The first step in this procedure, called "Chaos and Clues", is to determine whether or not a skin leison has something called "Chaos". We will use image recognition and machine learning to aid the dermatologist with this first algorithm-step, i.e., finding Chaos. As of now, this Chaos is devided into four classes: - Symmetrical colors, - Asymmetrical colors, - Symmetrical patterns, and - Asymmetrical patterns. We'll use a dataset of 20000 pictures and journal entries that will be processed and thinned down and put into a pre-trained convolutional neural network to a estimated size of 5000 pictures with a size of 400kB each.