Detection of diffuse abnormal perfusion in SPECT using a normal brain atlas.

Area of Science:

• Neuroimaging
• Medical image analysis
• Cerebrovascular disease detection

Background:

• Current visual assessment of Single Photon Emission Computerized Tomography (SPECT) cerebral perfusion studies suffers from significant inter- and intraobserver variability.
• Objective evaluation is needed to improve the diagnostic accuracy and consistency of SPECT perfusion analysis.

Purpose of the Study:

• To develop and validate an automated method for screening SPECT studies to detect diffuse abnormalities in cerebral blood flow.
• To establish a computerized atlas of normal regional cerebral blood flow (rCBF) for comparison.

Main Methods:

• Generation of a normal brain SPECT atlas by registering multiple normal studies, including mean intensity, nonlinear displacement, and activity pattern variance.
• Registration of patient SPECT volumes to the atlas to compute voxel-wise 3-D displacement and intensity differences.
• Classification of voxels as abnormal based on deviations exceeding 3 standard deviations from the normal mean.
• Validation using 24 abnormal and 21 normal SPECT perfusion studies, with Markovian segmentation for regional analysis.

Main Results:

• The automated method achieved an 80% correct classification rate for the whole brain.
• Classification accuracy increased to 93% when focusing on segmented gray matter regions.
• The method effectively identifies diffuse disseminated abnormalities in SPECT cerebral perfusion.

Conclusions:

• The developed automated method offers a reliable and objective approach for screening SPECT cerebral perfusion studies.
• This technique has the potential to reduce diagnostic variability and improve the detection of diffuse abnormalities.
• Automated analysis of SPECT scans using a computerized atlas can enhance diagnostic accuracy in neuroimaging.