This study investigates a simple Bayesian belief network for the diagnosis
of breast cancer, and specifically addresses the question of whether integr
ating image and non-image based features into a single network can yield be
tter performance than hybrid combinations of independent networks, From a d
ataset of 419 cases, including 92 malignancies, 13 features relating to mam
mographic findings, physical examinations and patients' clinical histories,
were extracted to build three Bayesian belief networks. The: scenarios tes
ted included a network incorporating: all features and two hybrids which co
mbined the outputs of sub-networks corresponding to the image or non-image
features. Average areas (A(z)) under the corresponding ROC curves were used
as measures of performance. The network incorporating only image based fea
tures performed better (A(z)=0.81) than that using nonimage features (A(z)
= 0.71). Both hybrid classifiers yielded better performance (A(z) = 0.85 fo
r averaging and A(z) = 0.87 for logistic regression), but neither hybrid wa
s as accurate as the network incorporating all features (A(z) = 0.89). This
preliminary study suggests that, like human observers who concurrently con
sider different types of information, a single classifier that simultaneous
ly evaluates both image and non-image information can achieve better diagno
stic performance than the hybrid combinations considered here. (C) 1999 Els
evier Science Ireland Ltd. All rights reserved.