Evaluation of a novel normal database with matched SPECT systems and optimal pre-filter parameters for 3D-SSP

Abstract

Purpose: A normal perfusion database (NDB) is imperative for the statistical imaging of brain function. This study validates a novel NDB created under the same injection dose and acquisition conditions for three gamma camera systems and evaluates optimal pre-filter parameters for three-dimensional stereotactic surface projections (3D-SSPs).

Method: We compared a novel NDB that matched the databases in each of three vendor gamma camera systems with a conventionally constructed NDB (conventional NDB) and a NDB constructed in-house for 3D-SSP. We generated hypoperfused regions where pre-specified volumes were simulated for various areas in SPECT images. The properties of each NDB were evaluated based on the distribution of the standard deviation (SD). Abnormal accumulation regions were validated using Z, extent, and artifactual scores. Detection error was used to evaluate the optimal Butterworth pre-filter cutoff frequency with the perfusion defect rate (PDR) in 3D-SSP.

Results: The SD distribution was the same in the novel NDB and in the NDB constructed in-house, and the SD of the peak distribution was 0.08-0.07. The Z and extent scores of the novel DB and the NDB constructed in-house were similar, but increased along with the artifactual scores when using the conventional NDB. Many artifacts appeared in the Z score map when using the conventional NDB. The detection error deviated from the actual value by -1.3% at a cutoff frequency of 0.58 cycles/cm and a PDR of 30%, which was the lowest. The cutoff frequency became lower or higher, and the low-perfusion defect rate increased according to the increasing detection error. The optimal cutoff frequency was between 0.52 and 0.58 cycles/cm.

Conclusions: We generated a novel NDB according to the individual devices and compared it with a conventional and a NDB constructed in-house. The Z and extent scores were essentially equal when using the novel DB and the NDB constructed in-house, but considerably differed when using the conventional NDB. The optimal cutoff frequency of the Butterworth filter evaluated from the detection error was in the range of 0.52-0.58 cycles/cm. The detection error increased the perfusion defect rate by <15% and this was undetectable in 3D-SSP. The next step will be to improve the accuracy of the extent of abnormal regions and the sensitivity of the Z score using a novel NDB constructed according to the individual devices.