Reduction of False Positives in Identification of Masses in Mammograms

Breast cancer has become one of the prevalent diseases in this era among women. Although breast cancer is a fatal disease, patients still have high chances of survival if detected at an early stage. The use of computer aided diagnosis (CAD) systems could provide an objective solution compared to qualitative solutions by radiologists. Many cases are getting resulted as false positives that the patients without cancer are sent to diagnostic mammography, on the other hand, patients with breast cancer are sent back saying that there is no cancer growth due to the inaccuracy which results in cancer enormously. In order to overcome this problem, a machine learning-based system is designed that can analyze the mammogram output and can predict whether the patient is developing cancer or not. To this end, a unique descriptor, namely Local Neighbourhood Intensity pattern (LNIP) is proposed for suspicious region representation. Experiments on DDSM database employing 2000 suspicious regions with support vector machine (SVM) classifier, demonstrate the efficiency of proposed system. The obtained result significantly reduces false positives achieving an accuracy of 97.5%.