FALSE-POSITIVE REDUCTION TECHNIQUE FOR DETECTION OF MASSES ON DIGITALMAMMOGRAMS - GLOBAL AND LOCAL MULTIRESOLUTION TEXTURE ANALYSIS

We investigated the application of multiresolution global and local te xture features to reduce false-positive detection in a computerized ma ss detection program. One hundred and sixty-eight digitized mammograms were randomly and equally divided into training and test groups. From these mammograms, two datasets were formed. The first dataset (manual ) contained four regions of interest (ROIs) selected manually from eac h of the mammograms. One of the four ROIs contained a biopsy-proven ma ss and the other three contained normal parenchyma, including dense, m ixed dense/fatty, and fatty tissues. The second dataset (hybrid) conta ined the manually extracted mass ROIs, along with normal tissue ROIs e xtracted by an automated Density-Weighted Contrast Enhancement (DWCE) algorithm as false-positive detections. A wavelet transform was used t o decompose an ROI into several scales. Global texture features were d erived from the low-pass coefficients in the wavelet transformed image s. Local texture features were calculated from the suspicious object a nd the peripheral subregions. Linear discriminant models using effecti ve features selected from the global, local, or combined feature space s were established to maximize the separation between masses and norma l tissue. Receiver Operating Characteristic (ROC) analysis was conduct ed to evaluate the classifier performance. The classification accuracy using global features were comparable to that using local features. W ith both global and local features, the average area, A,, under the te st ROC curve, reached 0.92 for the manual dataset and 0.96 for the hyb rid dataset, demonstrating statistically significant improvement over those obtained with global or local features alone. The results indica ted the effectiveness of the combined global and local features in the classification of masses and normal tissue for false-positive reducti on. (C) 1997 American Association of Physicists in Medicine.