Magnetic resonance (MR) imaging has been suggested as a technique for diagn
osing and monitoring myositis, an inflammatory muscle disease. To date, the
assessment of disease from MR images has been by subjective visual analysi
s. We describe here an objective, semi-automatic, computer-based method for
quantifying the degree of disease from MR images, without the need for a r
adiologist or physician trained in the visual assessment of the MR images.
The method is based on analysis of the histogram of intensity values produc
ed from the MR images. The analysis yielded measures of the intensity and e
xtent of disease. These two measures were combined to produce a calculated
myositis index (CMI) which described the degree of disease evident from the
MR images, This index was compared with a clinical assessment of the patie
nt's condition, based on currently accepted, invasive and non-invasive, non
-imaging criteria. Receiver operating characteristic (ROC) curve analysis s
howed that calculated myositis index agreed at least as well with clinical
assessment as did visual analysis (receiver operating characteristic area =
0.93 and 0.94, p = not significant (NS), respectively, for separating remi
ssion from disease). Even using only two central MR slices for each patient
, the receiver operating characteristic area for calculated myositis index
was 0.92, implying that very short acquisition times are possible. We concl
ude that quantitative histogram analysis of MR images can be successfully p
erformed with minimal operator input and using few MR slices. Agreement wit
h more invasive clinical assessment is good and the method has the advantag
es of repeatability, objectivity, and decreased scan and analysis time. (C)
1999 Elsevier Science Inc.