Paper Details: Downloads: 1563
Serial Number: P1151637518
Title: Automatic Recognition of Lesion-like Regions in Black Skin Medical Images
Authors: Géraud Azehoun-Pazou and Kokou Assogba
Abstract: This paper presents a study made to automatically recognize lesion-like regions in black skin medical images. Skin lesions have consistently had one of the most rapidly increasing incidences of all cancers. Early diagnosis is particularly important but it is a challenging task, especially for black populations. Moreover, black skin specialists are often limited to the use of classic macroscopic images for diagnosis purpose. All these difficulties are related to black skin pigmentation level and the small visual differences between lesion parts and healthy ones. We propose here a computerized method which identifies automatically lesion regions with more accuracy and efficiency. It works like an automaton that traverse lesion images and automatically classifies healthy regions from lesions regions. The designed classifier is a Multi-Layer Perceptron Artificial Neural Network (MLP-ANN) trained with color and texture features. We made many combinations of features and varied the number of neurons in hidden layer, in order to obtain best performances. Nine features (six of texture and three of color) have been retained to train the network. The achieved classification performance is 97.2% in both training, validation and testing set.
Keywords: Medical images, Neural Network, Lesion recognition, Black skin, Texture features
Journal/Conference: International Journal of Graphics, Vision and Image Processing
Volume: 16
Issue: 3
Submission Date: 9/7/2016 12:00:00 AM
Review Date: 9/21/2016 12:00:00 AM
Publishing Date: 10/27/2016 12:00:00 AM
Article Downloads: 1563
Download:

Facebook