Information Technology Reference
In-Depth Information
facilitate clinicians as intelligent assistants in diagnostic and prognostic pro-
cesses, laboratory analysis, treatment protocol, and teaching of medical students.
The initial research efforts in this area were devoted to the development of
medical expert systems . Their main components comprise knowledge base, con-
taining specific facts from the application area, rule-based system to extract the
knowledge, explanation module to support and explain the decisions made to
the end user, and user interface to provide natural communication between the
system and the human expert. One of the early prominent examples is MYCIN,
developed in mid-1970s to diagnose and treat patients with infectious blood dis-
eases caused by bacteria in the blood [6]. Up to date, the number of developed
medical expert systems has grown up enormously as discussed in [7].
With the extensive use of diagnostics imaging techniques such as X-ray, Mag-
netic resonance imaging (MRI) and Ultrasound, recently, another promising type
of medical intelligent paradigm has emerged, namely computer-aided detection ,
or shortly CAD . As its name implies, the primary goal is to assist, rather than
to substitute, the human expert in image analysis tasks, allowing him compre-
hensive evaluation in a short time. One typical application area of CAD systems
is the breast cancer detection and diagnosis. In [8], it is presented a general
overview of various tasks such as image feature extraction, classification of mi-
crocalcifications and masses, and prognosis of breast cancer as well as the related
techniques to tackle these tasks, including Bayesian networks, neural networks,
statistical and Cox prognostic models. Next, we review some of the recent tech-
niques used for the development of breast cancer CAD systems, and then we
discuss a number of previous studies dealing with multi-view dependencies to
facilitate mammographic analysis.
3.1 Computer-Aided Detection for Breast Cancer
In some previous research, Bayesian network technology has been used in or-
der to model the intrinsic uncertainty of the breast cancer domain [9,10]. Such
models incorporate background knowledge in terms of (causal) relations among
variables. However, they use BI-RADS terms to describe a lesion, rather than
numerical features automatically extracted from images. This requires the hu-
man expert to define and provide the input features a priori, which limits the
automatic support of the system. Other research attempt is the expert system
for breast cancer detection, proposed in [11]. The main techniques used are as-
sociation rules for reducing the feature space and neural networks for classifying
the cases. Application on the Wisconsin breast cancer dataset [12] showed that
the proposed synthesis approach has the advantage of reducing the feature space,
allowing for better discrimination between cancerous and normal cases. A limita-
tion of this study, however, is the lack of any clinically relevant application of the
proposed method with a larger dataset and an extensive validation procedure.
Automatic extraction of features is applied in [13], where the authors propose
a CAD system for automatic detection of miscrocalcifications on mammograms.
The method uses least-squares support vector machines to fit the intensity sur-
face of the digital X-ray image and then convolution of the fitted image with
 
Search WWH ::




Custom Search