Evaluation of surface roughness of tumor using neural network

Since surface roughness of a malignant tumor is more remarkable than that of a benign tumor, it is possible to classify pathological states of the tumor by computing the degree of surface roughness. We have proposed a method for the segmentation of the tumor in ultrasonic echography and confirmed the feasibility of the technique. This paper describes a neural network based classifier using the surface roughness of a breast tumor which is extracted from ultrasonic echography. We define nine parameters for evaluation of the surface roughness, which form an artificial neural network (ANN) input vector. The ANN output sequence displays two types of pathological states; malignant and benign. We use twenty seven benign tumors and twenty four malignant tumors for the feasibility study. Twenty four tumors are used for learning data and the other tumors are used for the trial. As a result, successful classification is obtained