A Toolbox for Analytical Detection and Sequential Evaluation of Small Brain Metastasis in Stereotactic Radiosurgery

Abstract

Present a toolbox using high resolution MRI brain scans to (A) analytically detect initial small metastasis in stereotactic radiosurgery patients, (B) enhance detection of new metastasis in follow-up studies, and (C) consistently compute volume changes. MPR transaxial MRI scans are acquired using identical protocols for all data sets. Raster scans of unprocessed individual images and pertinent DICOM header information are transferred to the toolbox. A case study was performed for each toolbox component A. The algorithm used to initially detect very small metastasis incorporates several clinical rules but most notable is the identification of a single pixel volume outside of normal high density region such as bone that is not present on adjacent slices. The very small lesion (VSL) in this study was specifically characterized by: (1) a single pixel that is significantly higher than the adjacent pixels, (2) the pixel intensity is in a select range above normal tissue but below higher intensities in normal structures or contrast, and (3) the high intensity is only present on one transaxial slice. B. The MRI for the first treatment (Primary) and the repeat MRI for the second treatment (Secondary) were fused using Cross Correlation. Features such as user selectable display scale, profiling, and on demand views of individual or fused data sets are available. Identical display Window and Level were then set for both Primary and Secondary such that the known targets for the first treatment were clearly demonstrated. Since the brain is a relatively rigid structure and the images are usually less than three months apart with nearly identical imaging parameters, normal background structures can be easily minimized by simply subtracting a fraction of the Primary from the Secondary. Further enhancement is achieved by pixel renormalization. C. Five data sets encompassing 450 days post radiosurgery were analyzed on a representative patient. Each follow-up scan was fused to the initial treatment scan. Automatic target contours were generated on all data sets using a set threshold above adjacent normal tissue. The computed volumes versus days after treatment were then plotted. A. The algorithm automatically identified regions with potential small metastasis that were approved by an experienced user; a minimal false positive rate was observed. B. The enhanced images tend to minimize areas with similar features and accentuate areas with differences. A fifty percent increase in target to surrounding normal structure intensity ratio was achieved. C. The initial target volume increased fifty percent within ninety days then decreased linearly to zero at 370 days post treatment. The combined algorithms assist in speedy throughput for initial and retreatment, eliminate some observer dependence, and are useful in retrospective reviews with multiple data sets.