IPES: An Image Processing-Enabled Expert System for the Detection of Breast Malignant Tumors

Abstract

Mammography is an influential screening tool for preliminary detection of breast carcinoma. However, Interpreting mammograms is exhausting, particularly in the screening context. Moreover, sensitivity of mammography based screening is influenced by image quality and the experience level of radiologist. Thus, computer-aided diagnosis (CAD) programs can be utilized as a second-opinion tools that enhance the performance of radiologists, by consolidating sensitivity rates in contrast with those taken by double readings. The current paper is directed towards the integration of image processing and the rule-based reasoning into a diagnostic expert system for breast tumors. A proposed system termed IPES (Image Processing-enabled Expert System) is developed for the detection of breast malignant tumors, appear in mammography in three steps: (1) segmentation of mammographic masses from both pectoral muscle region and breast tissues, (2) Characterization of segmented masses upon the standards of Breast Imaging Reporting and Data System (BI-RADS) by: (i) shape-relevant features; (ii) margin characteristics; and (iii) density features, and (3) diagnosis of mass type upon some inference rules. The data set used for testing IPES contained 540 samples obtained from the standard Digital Database for Screening Mammography (DDSM). The Receiver Operator Characteristic (ROC) curves have been employed to evaluate the sensitivities and specificities of the system. Finally, the results reveal the efficacy of IBES in discriminating both malignant and benign breast masses.