A computer aid for the detection of suspicious microcalcification clusters in digitized mammograms

This work focused on the development of a computer-aided diagnosis system for the automatic detection of microcalcification clusters in mammograms. The objective of this study was to develop an accurate computer aid and show its potential as a clinical tool towards the early detection of breast cancer.; The design and evaluation of the algorithm involved three main phases. In the first phase of the algorithm, a method was developed for the automated detection of microcalcification clusters. Method development involved extracting histogram features describing individual microcalcifications and cluster features describing microcalcification clusters and designing rule-based classifiers incorporating these features. A database of 98 images was used during this phase for adjusting the parameters and for initial assessment of the method. For the first phase, parameters of the algorithm were adjusted manually since the objective was to examine the discriminating ability of the features towards the accurate detection of microcalcification clusters.; In the second phase of the algorithm, a method for automatic parameter optimization of the CAD scheme was developed. The objectives of the second phase were: (a) to make the training of the algorithm practical, (b) to avoid dependence of the algorithm on subjective rules that might not generalize on a broader population, and (c) to allow the performance evaluation of the algorithm using FROC analysis. The result of the second phase was a neural network-based algorithm.; In the third and last phase of the algorithm, the neural network-based algorithm was retrained and evaluated on an independent, publicly available database to examine the ability of the algorithm to generalize its performance on an unknown population and to enable comparisons with methods from different laboratories.; At the conclusion of this work, the algorithm presented herein represents a complete, automated CAD system. The system was designed to allow relatively straightforward re-optimization for different data sets and was carefully evaluated for its ability to generalize to an independent, large, publicly available data set. The performance of the CAD system supports its application as an accurate computer aid for the detection of microcalcification clusters in mammograms.