This paper introduces a number of reliability criteria for computer-aided d
iagnostic systems for breast cancer. These criteria are then used to analyz
e some published neural network systems. It is also shown that the property
of monotonicity for the data is rather natural in this medical domain, and
it has the potential to significantly improve the reliability of breast ca
ncer diagnosis while maintaining a general representation pou er. A central
part of this paper is devoted to the representation/narrow vicinity hypoth
esis, upon which existing computer-aided diagnostic methods heavily rely. T
he paper also develops a framework for determining the validity of this hyp
othesis. The same framework can be used to construct a diagnostic procedure
with improved reliability. (C) 2000 Academic Press.