Texture Classification using Artificial Neural Network for Diagnosis of Skin Cancer

Dalia N. Abdul-Wadood; Dr. Loay E. George; Dr. Nabeel A. Rasheed

This paper attempts to improve the efficiency of the system that proposed in [1] to determine whether a given skin lesion microscopic image is malignant or benign; in case of malignancy, the system can specify its type; whether it is squamous cell carcinoma or basal cell carcinoma (the two leading skin cancer types). The testing of this system was conducted using 80 microscopic images of skin tissues of the types normal, benign and the two types of skin cancer (squamous and basal); the images have been collected from different hospital pathology departments as part of the research work. Some of the collected samples have been used as training and others as testing materials. The proposed system consists of 3 main steps. First, extraction of a set of textural descriptors to localize the abnormal visual attributes which may appear in the tested skin tissue images. Second, selection of the best discriminating texture features. Third, identify the type of skin tissue images using artificial neural network (ANN). In the training phase, the system was trained using 50 skin tissue images, the textural features extracted from training samples were analyzed and their discrimination powers were evaluated in order to get a list of the most suitable features for auto recognition task. When ANN is trained on co-occurrence features the attained allocation accuracy rates was (%97.71) and the diagnosis accuracy rate was (%98.75). While when using ANN with combinations of different types of textural features; the allocation accuracy rate reached to (%97.90) while the diagnosis accuracy rate became (%98.75).

Select Volume / Issues:

• Volume 5,Issue 4, July 2014

Year:

2014

Type of Publication:

Article

Keywords:

Medical Image Classification; Skin Cancer Detection; Medical Image Analysis; Textural Analysis; Artificial Neural Network

Journal:

IJECCE

Volume:

5

Number:

4

Pages:

836-841

Month:

July