MammoSIGHT
Missed cancer reading in mammography is a major source of lawsuits in radiology. Studies have shown that among newly diagnosed breast cancer cases in which the patients have previous mammograms, 75% of the cases will have abnormality detectable in the old films.
It is also known that a very effective way to reduce the errors is to have two radiologists read the same mammograms independently. However, it is practically infeasible to implement such a two-radiologist reading system. A computer system serving as a second reader is therefore an attractive option.
Detecting malignancy in mammograms can be difficult. Everyone’s mammogram is unique and there can be a great deal of variations among “normal” images. Unlike CT and MRI, mammograms are not cross sectional images. The projection from 3D to 2D and the resulting overlaps on the images may interfere with the recognition of the distinguishing features. The features are often very subtle. The rules for differentiating the benign and malignant cases are vague and not easily formulated.
In view of the nature of this problem, a machine leaning (SVM) based approach is a natural choice for implementing such a system.
The illustration below shows three SVM based subsystems are developed separately. The basic structures of the subsystems are similar.
The detection component finds the areas of particular interest in the image and separate the objects from the background.
The feature extraction component formulates numerical values relevant to the classification task from the segmented objects.
The SVM classifier produces an index discriminating between the benign and malignant cases.
The individual components can be developed in parallel because of the modular structure. In developing the calcification segmentation component a selected set of malignant, benign, and normal cases representing a wide range of images was used to guide and test the design in order to produce a general, robust and accurate algorithm. At the same time, the SVM classifier was developed and tested with manually prepared input data. A set of 300 images (150 benign and 150 malignant cases) was used in training the SVM. An independent set of 328 images was used for testing. High dimensional input features were used to ensure a sufficient capacity for automatically extracted features. The components will be integrated and adjusted for optimal performance.
Calcification detection and segmentation
Clusters of micro calcifications are characterized by their relatively small sizes and high densities. The algorithm combines a recursive peak seeking technique with morphological operations to achieve a highly accurate calcification detection and segmentation.
|
