An Advanced Probabilistic Framework for Assisting Screening Mammogram Interpretation - Computational Intelligence in Healthcare: Advanced Methodologies - page 359

by a CAD system) or false positive; see Figure 1. A region detected by a CAD
system is described by a number of continuous (real-valued) features describing,
for example, size, density and location of the region. By link we denote matching
(established correspondence) between two regions in MLO and CC views. The
term case refers to a patient who has undergone a mammographic exam.
In addition, two main types of mammographic abnormalities are distinguished:
microcalcifications and masses. Microcalcifications are tiny deposits of calcium
and are associated with extra cell activity in breast tissue. They can be scat-
tered throughout the mammary gland, which usually is a benign sign, or occur
in clusters, which might indicate breast cancer at an early stage.
According to the BI-RADS [3] definition, “a mass is a space occupying le-
sion seen in two different projections.” When visible in only one projection, it
is referred as a “mammographic density”. However, the density may be a mass,
perhaps obscured by overlying glandular tissue on the other view, and if it is
characterised by enough malignant features then the patient would be referred
for further examination. For instance, many cancerous tumours may be pre-
sented as relatively circumscribed masses. Some might be sharply delineated or
may be partly irregular in outline, while other cancers (which induce a reactive
fibrosis) have an irregular or stellated appearance resulting in a characteristic
radiographic feature, so-called spiculation. Masses might vary in size, usually
not exceeding 5 cm. In terms of density, they appear denser (brighter on the
mammogram) when compared to breast fat tissue.
Masses are the more frequently occurring mammographic type of breast can-
cer. In the context of screening mammography, there is strong evidence that
misinterpretation of masses is a more common cause of missing cancer than per-
ceptual oversight. Furthermore, applications have shown that CAD tends to per-
form better in identifying malignant microcalcifications compared with masses
[4]. Masses are more di cult to detect due to the great variability in their phys-
ical features and similarity to the breast tissue, especially at early stages of
development. Hence, the prompt of the current CAD comprises not only the
cancer but also a large number of false positive (FP) locations-undesired result
in screening where the reading time is crucial. Given the challenge in true mass
identification, in this work, we focus only on the detection of malignant masses.
3 Previous Research
The rapid development of computer technology in the last three decades has
brought unique opportunities for building up and employing numerous auto-
mated intelligent systems in order to facilitate human experts in decision making,
to increase human's performance and eciency, and to allow cost saving. One
of the domains where intelligent systems have been applied with great success is
healthcare [5]. Healthcare practitioners usually deal with large amounts of infor-
mation ranging from patient and demographic to clinical and billing data. These
data need to be processed eciently and correctly, which often creates enormous
pressure for the human experts. Hence, healthcare systems have emerged to