The automated detection of clusters of microcalcifications

The algorithm described demonstrates the feasibility of automated detection of microcalcification clusters using edge detection, graph searching, linear discriminant analysis and cluster detection techniques. It represents a significant improvement over other methods with 100% true-positives, 0% false positives using the re-substitution method and 98% true-positives, 0% false positives using the leave-one-out method. The leave-one-out method resulted in one abnormal region being misclassified. The error occurred not in the detection of potential microcalcifications, but in their classification prior to cluster detection. A number of possible reasons for this misclassification are given.<
>