22-05-2014, 04:52 PM
A Comparative Study on Microcalcification Detection Methods with Posterior Probability Estimation based on Gaussian Mixture Models
A Comparative Study on Microcalcification.pdf (Size: 280.39 KB / Downloads: 17)
Abstract
Automatic detection of microcalcifications in
mammograms constitutes a helpful tool in breast cancer diagno-
sis. Radiologist’s confidence level on microcalcification detection
would be improved if a probability estimate of its presence could
be obtained from computer-aided diagnosis. In this paper we
explore detection performance of a simple Bayesian classifier
based on Gaussian mixture probability density functions (pdf).
Posterior probability of microcalcification presence may be
estimated from the probabilistic model. Two model selection
algorithms have been tested, one based on the Minimum
Message Length criterion and the other on discriminative
criteria obtained from the classifier performance. In addition,
we propose a complementing model selection algorithm in order
to improve the initial system performance obtained with these
methods. Simulation results show that our model gets a good
compromise between classification performance and probability
estimation accuracy.
I NTRODUCTION
Computer-aided detection (CAD) of breast carcinomas is
a key tool for lesion identification improvement in mam-
mograms (10% increase in radiologist’s sensitivity [1]).
Formulation of automated microcalcification detection under
a statistical pattern recognition framework, allows to take
some advantages from the probabilistic theory. If a proper
probability density function model could be extracted from
the feature space, the posterior probability of a microcalci-
fication to appear in a specific ROI (region of interest) of
a digital image could be estimated as well. This would end
up in an automated system that, apart from taking decisions,
might give a confidence level on taking those decisions.
C ONCLUSIONS
In this paper, we have evaluated a Bayesian classifier
based on Gaussian mixture models for microcalcification
detection and posterior probability estimation. Two model
selection methods have been evaluated in terms of detection
and estimation performance, with results in line with those
presented in [2] (Table 4). Additionally, we have proposed
an additional model selection algorithm based on prune, split
and merge operations, and showed that its application to
a model previously initialized with the GGMNu algorithm
gives a good compromise between entropic cost and area
under ROC curve.