MED-NODE: A computer-assisted melanoma diagnosis system using non-dermoscopic images

•Melanoma is an aggressive type of skin difficult to differentiate from benign naevi.•We present an expert system that assists physicians with this task.•The system uses color and texture lesion descriptors from non-dermoscopic images.•The system also uses a set of visual attributes provided by the...

Full description

Saved in:
Bibliographic Details
Published in:Expert systems with applications 2015-11, Vol.42 (19), p.6578-6585
Main Authors: Giotis, Ioannis, Molders, Nynke, Land, Sander, Biehl, Michael, Jonkman, Marcel F., Petkov, Nicolai
Format: Article
Language:English
Subjects:
Automation
Classification
Computer-assisted diagnosis
Dermatology
Diagnostic software
Lesions
Medical imaging
Melanoma
Modular
Physicians
Surface layer
Texture
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:•Melanoma is an aggressive type of skin difficult to differentiate from benign naevi.•We present an expert system that assists physicians with this task.•The system uses color and texture lesion descriptors from non-dermoscopic images.•The system also uses a set of visual attributes provided by the examining physician.•The proposed system achieves high diagnostic accuracy results (81%). Melanoma is one of the most aggressive types of skin cancer and in many cases it is difficult to differentiate from benign naevi. In this contribution we present a decision support (expert) system, which we call MED-NODE, able to assist physicians with this challenging task. The proposed system makes use of non-dermoscopic digital images of lesions from which it automatically extracts the lesion regions and then computes descriptors regarding the color and texture. In addition, a set of visual attributes is provided by the examining physician. The automatically extracted descriptors and the attributes provided by the physician are separately used for automatic prediction. Final classification is achieved by a majority vote of all predictions. The proposed system achieves high diagnostic accuracy results (81%) and performs comparably to state-of-the-art methods that are using dermoscopic images, though such images contain more detailed information and are subject to less noise and illumination effects. The simple input requirements and the robustness of its descriptors allow MED-NODE to be an effective tool within the diagnostic process for melanoma. In addition, the modular nature of the system allows for it to be easily extended.
ISSN:0957-4174
1873-6793
DOI:10.1016/j.eswa.2015.04.034