Neural Network Analysis of Breast Cancer from Mammography Findings

Introduction: About 240,000 women are diagnosed with breast cancer each year. [14] Considering the growth of breast cancer in Iran and decreasing the age of patients are required to use of diagnostic procedures with high accuracy. Patients referred to biopsy, are faced with financial, physical, and mental problems. So it is needed the methods which have capability to diagnose the cancer noninvasively. [2], [15], [16], [17] The aim of this work is designation of an Artificial Neural Network (ANN) model to detect the malignity agent from mammography. It is expected such an ANN is capable to distinguish benign and malignant tumor with high accuracy without using biopsy findings.Method: The mammography findings belong to 102 patients have been referred to one of the busy cancer research centers. There are 5 important features in radiologist reports including: micro calcification, density, lymph node, nipple retraction and lymphadenopathy. The model collected data from 87 patients used for training stage that are given to the model as an input matrix. Then the model tested the program with data from 15 patients that are given as a set of testing data. These 2 sets of data are selected randomly by the operator. Training stage has to be done via testing data and considering those 5 features and the level of their importance. The level of importance is computed via its repetition among 87 patients and demonstrated as a weight of each feature. Then it is expected to use these weights to predict the result of biopsy. [4], [6], [7], [8], [9], [10], [11], [12], [13] Result: The average accuracy, specificity and sensitivity of the proposed model are equal to 86.66%, 77.8%, and 91.66%, respectively. The results are compared with the values obtained by radiologist predicting that are equal to 80%, 87.5%, and 71.42%, respectively. [1], [7], [18], [19]Conclusion: The ANN which is used in this study, demonstrated higher accuracy and sensitivity for predicting results of biopsy comparing predicting result obtained by the radiologist.

Keywords:

breast cancer , biopsy , micro calcification , ANN