Malignant densities in mammograms have an irregular appearance and frequently are surrounded by a radiating pattern of linear spicules. In this paper a method is described to detect such stellate patterns. This method is based on statistical analysis of a map of pixel orientations. If an increase of pixels pointing to a region is found, this region is marked as suspicious, especially if such an increase is found in many directions. Orientations of the image intensity map are determined at each pixel using a multiscale approach. At a given scale, accurate line-based orientation estimates are obtained from the output of three-directional, second-order, Gaussian derivative operators. The orientation at the scale at which these operators have maximum response is selected. If a line-like structure is present at a given site, this method provides an estimate of the orientation of this structure, whereas in other cases the image noise will generate a random orientation. The pixel orientation map is used to construct two operators which are sensitive to radial patterns of straight lines. Combination of the output of these operators using a classifier allows for detection of stellate patterns. Different classification methods have been compared and results obtained on a common database are presented. Around 90% of the malignant cases were detected at rate of one false positive (FP) per image.

Specificity levels of automatic mass detection methods in mammography are generally rather low, because suspicious looking normal tissue is often hard to discriminate from real malignant masses. In this work a number of features were defined that are related to image characteristics that radiologists use to discriminate real lesions from normal tissue. An artificial neural network was used to map the computed features to a measure of suspiciousness for each region that was found suspicious by a mass detection method. Two data sets were used to test the method. The first set of 72 malignant cases (132 films) was a consecutive series taken from the Nijmegen screening programme, 208 normal films were added to improve the estimation of the specificity of the method. The second set was part of the new DDSM data set from the University of South Florida. A total of 193 cases (772 films) with 372 annotated malignancies was used. The measure of suspiciousness that was computed using the image characteristics was successful in discriminating tumours from false positive detections. Approximately 75% of all cancers were detected in at least one view at a specificity level of 0.1 false positive per image.

Scale is an important issue in the automated detection of masses in mammograms, due to the range of possible sizes masses can have. In this work, it was examined if detection of masses can be done at a single scale, or whether it is more appropriate to use the output of the detection method at different scales in a multiscale scheme. Three different pixel-based mass-detection methods were used for this purpose. The first method is based on convolution of a mammogram with the Laplacian of a Gaussian, the second method is based on correlation with a model of a mass, and the third is a new approach, based on statistical analysis of gradient-orientation maps. Experiments with simulated masses indicated that little can be gained by applying the methods at a number of scales. These results were confirmed by experiments on a set of 71 cases (132 mammograms) containing a malignant tumor. The performance of each method in a multiscale scheme was similar to the performance at the optimal single scale. A slight improvement was found for the correlation method when the output of different scales was combined. This was especially evident at low specificity levels. The correlation method and the gradient-orientation-analysis method have similar performances. A sensitivity of approximately 75% is reached at a level of one false positive per image. The method based on convolution with the Laplacian of the Gaussian performed considerably worse, in both a single and multiscale scheme.

State-of-the-art algorithms for detection of masses in mammograms are very sensitive but they also detect many normal regions with slightly suspicious features. Based on segmentations of detected regions, shape and intensity features can be computed that discriminate between normal and abnormal regions. These features can be used to discard false positive detections and hence improve the specificity of the detection method. In this work two different methods to segment suspect regions were examined. A number of different implementations of a region growing method were compared to a discrete dynamic contour method. Both methods were applied to a consecutive data set of 132 mammograms containing masses and architectural distortions, taken from the Dutch screening program. Evaluation of the performance of the methods was done in two different ways. In the first experiment, the segmentations of masses were compared to annotations made by the radiologist. In the second experiment, a number of features were computed for all segmented areas, normal and abnormal, based on which regions were classified with a neural network. The most sophisticated region growing method and the method using the dynamic contour model had a similar performance when evaluation was based on the overlap of the annotations. The second experiment showed that the contours generated by the discrete dynamic contour model were more suited for computation of discriminating features. Contrast features were especially useful to improve the performance of the detection method.

A substantial proportion of cancers that were missed in a screening program, despite double reading, were found with this detection method at less than one false-positive finding per image.