Volume determination in cerebral tumors requires accurate and reproduc
ible segmentation. This task has been traditionally accomplished using
planimetric methods which define the boundary of the lesion using thr
esholding and edge detection schemes. These methods lack accuracy and
reproducibility when the contrast between the lesion and surrounding t
issue is not maximized. Because of this limitation contrast agents are
used providing reproducible results for the enhancing portion of the
lesion. A novel approach for volume determination has been developed (
eigenimage filter) which segments a desired feature (tissue type) from
surrounding undesired features in a sequence of images. This method c
orrects for partial volume effects and has been shown to provide accur
ate and reproducible volume determinations. In addition, the eigenimag
e filter does not require the use of contrast and has the capability t
o segment a lesion into multiple regions. This allows different compon
ents of the lesion to be included and monitored in treatment. In this
study planimetric methods and the eigenimage filter were compared for
segmenting cerebral tumors and determining their volumes. The planimet
ric methods were reproducible in determining volumes for the enhancing
portion of the lesion with interobserver percent differences < 8% and
intraobserver percent differences < 4%. The eigenimage filter had int
erobserver percent differences < 7% and intraobserver percent differen
ces < 3%. In the eigenimage procedure both the enhancing portion of th
e lesion as well as additional regions within the lesion were identifi
ed. Comparing the results obtained from the two methods demonstrated g
ood agreement for presurgical studies (percent differences < 9%). When
comparing postsurgical studies large differences were seen. In the po
stsurgical studies the eigenimage method allowed multiple regions to b
e followed in subsequent MRI and in two patients showed a volume chang
e that suggested tumor recurrence more clearly. Since the amount of in
formation obtained using the eigenimage filter may allow a more comple
te assessment of the lesion, it is suggested that it could improve the
clinical evaluation of cerebral tumors. (C) 1996 American Association
of Physicists in Medicine.