Improved delineation of brain tumors: an automated method for segmentation based on pathologic changes of 1H-MRSI...

In this study, we developed a method to improve the delineation of intrinsic brain tumors based on the changes in metabolism due to tumor infiltration. Proton magnetic resonance spectroscopic imaging ((1)H-MRSI) with a nominal voxel size of 0.45 cm(3) was used to investigate the spatial distribution of choline-containing compounds (Cho), creatine (Cr) and N-acetyl-aspartate (NAA) in brain tumors and normal brain. Ten patients with untreated gliomas were examined on a 1.5 T clinical scanner using a MRSI sequence with PRESS volume preselection. Metabolic maps of Cho, Cr, NAA and Cho/NAA ratios were calculated. Tumors were automatically segmented in the Cho/NAA images based on the assumption of Gaussian distribution of Cho/NAA values in normal brain using a limit for normal brain tissue of the mean + three times the standard deviation. Based on this threshold, an area was calculated which was delineated as pathologic tissue. This area was then compared to areas of hyperintense signal caused by the tumor in T2-weighted MRI, which were determined by a region growing algorithm in combination with visual inspection by two experienced clinicians. The area that was abnormal on (1)H-MRSI exceeded the area delineated via T2 signal changes in the tumor (mean difference 24%) in all cases. For verification of higher sensitivity of our spectroscopic imaging strategy we developed a method for coregistration of MRI and MRSI data sets. Integration of the biochemical information into a frameless stereotactic system allowed biopsy sampling from the brain areas that showed normal T2-weighted signal but abnormal (1)H-MRSI changes. The histological findings showed tumor infiltration ranging from about 4-17% in areas differentiated from normal tissue by (1)H-MRSI only. We conclude that high spatial resolution (1)H-MRSI (nominal voxel size = 0.45 cm(3)) in combination with our segmentation algorithm can improve delineation of tumor borders compared to routine MRI tumor diagnosis.