TY - JOUR T1 - Automated Detection of Local Normalization Areas for Ictal-Interictal Subtraction Brain SPECT JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 1419 LP - 1425 VL - 43 IS - 11 AU - Nicolas Boussion AU - Claire Houzard AU - Karine Ostrowsky AU - Philippe Ryvlin AU - François Mauguière AU - Luc Cinotti Y1 - 2002/11/01 UR - http://jnm.snmjournals.org/content/43/11/1419.abstract N2 - Whole-brain activity is often chosen to quantitatively normalize peri-ictal and interictal SPECT scans before their subtraction. This use is not justified, because significant and extended modification of the cerebral blood flow can occur during a seizure. We validated and compared 2 automatic methods able to determine the optimal reference region, using simulation and clinical data. Methods: In the first method, the selected reference region is the intersection of peri-ictal-interictal areas with no significantly different z values. The other method relies on a 3-dimensional iterative voxel aggregation. The increase of the selected volume is stopped by using 2 different variance tests (Levene and SE). These algorithms were tested on 39 epileptic patients and were validated using 1 interictal and 10 peri-ictal scans simulated from the mean image of 22 healthy subjects. Results: In the patient studies, the mean relative activity of the selected regions, compared with whole-brain activity (classic normalization), was 122.6%. Their average relative size (compared with the size of the whole brain) was 33.2% for the z map method, 22.8% for the SE test, and 11.8% for the Levene test. After application of our automatic processes, subtraction of the simulated images revealed a recovery of abnormal regions up to 45% larger than the region obtained with classic normalization. Conclusion: These results illustrate the role of normalization on the subtracted peri-ictal and interictal images. Our methods are automatic and objective and give good results on various simulated images. The z map construction is worth considering because it is simple, selects large parts of the brain, and requires little computation time. ER -