Mahidol University Logo
Faculty of ICT, Mahidol University
 

Admissions

Printable Version

 

CLASSIFICATION OF MAMMOGRAPHIC MASSES USING 3D SURFACE TEXTURE

 

TITLE CLASSIFICATION OF MAMMOGRAPHIC MASSES USING 3D SURFACE TEXTURE.
AUTHOR THANAPOL INTRAVESN
DEGREE MASTER OF SCIENCE PROGRAMME IN COMPUTER SCIENCE
FACULTY FACULTY OF SCIENCE
ADVISOR SUKANYA PHONGSUPHAP
CO-ADVISOR CHOMTIP PRONPANOMCHAI
 
ABSTRACT
Breast cancer is still the cause of death of several millions of women around the world. In medical fields, X-ray mammogram or breast X-ray is a popular examination method for detection of breast cancer because it is the most cost-effective and accurate method of early detection of breast cancer. Nevertheless, radiologists encounter problems of misdiagnosis which may be cause by inexperience. This problem can be decreased by using computer-aided diagnosis (CAD). In the past few years, several researchers have investigated the problem of computer-aided diagnosis via digital mammograms. The primary features used to classify mammographic masses relate masses' size, shape, margin, density, texture features and so on. This research proposed an approach of classifying mammographic mass images using 3D surface texture features, focusing on mammographic mass associated with calcifications (calcified mammographic mass) only. The 3D texture features were applied to improve the performance of traditional texture features consisting of the roughness average (Ra), roughness root mean square (Rq), skewness (Rsk) and kurtosis (Rku). In previous research, it was reported that the effective traditional texture features based on spatial gray-level co-occurrence matrices (SGCMs) consisted of inverse difference moment (distance = 1 at direction = 0? and 135?) and correlation (distance = 1 at direction = 135?). Thus in this study, the performance of using 3D surface texture features was compared with that of using the traditional texture features based on SGCMs. The sample dataset involved 100 digital calcified mammographic masses, which consisted of 50 benign and 50 malignant images from the Mammographic Image Analysis Society (MIAS). This sample dataset was a standard dataset for the research in mammographic image analysis. The experiments in this research conducted five different classifiers: Euclidean distance, Mahalanobis distance and three different structures of Multi-layer Perceptron Neural Network (MPNN) classifiers. It was found that the best classification accuracy rate achieved was the method based on the Neural Network with two hidden layers, each with five nodes, using a combination of 3D surface texture features and traditional texture features. It achieved the highest classification accuracy rate of 76% with the true positive rate (TP) of 0.72 and the true negative rate (TN) of 0.80. Using 3D surface texture features alone, it attained classification accuracy rate of 70% with the TP rate of 0.68 and the TN rate of 0.72. Finally, using traditional texture features achieved the lowest classification accuracy rate of 68% with the TP rate of 0.64 and the TN rate of 0.72.
KEYWORD MAMMOGRAM CLASSIFICATION / CALCIFIED MAMMOGRAPHIC MASSES / 3D SURFACE TEXTURES / NEURAL NETWORKS

 

Go to Top

 

ICT Building, Mahidol University, 999 Phuttamonthon 4 Road, Salaya, Nakhonpathom 73170 Tel. +66 02 441-0909 Fax. +66 02 849-6099
Mahidol University Computing Center, The Faculty of ICT, Mahidol University , Rama 6 Road, Rajathevi, Bangkok 10400 Tel. +66 02 354-4333 Fax. +66 02 354-7333