Investigating the Impact of Imaging Parameters on Quantification and Interpretation in Brain PET

Abstract

Rationale CortexID software (GE Healthcare) assists in interpreting and assessing FDG and beta-amyloid brain Positron Emission Tomography (PET) images for neurological disorders. It provides Z-scores that indicate abnormal radioactive tracer uptake. These Z-scores are based on the patient’s standardized up take value ratios (SUVRs) compared to a chosen reference brain region. They reflect deviations from SUVRs in healthy individuals of the same age. A positive Z-score indicates an increased uptake and a negative Z-score indicates a decreased uptake of the radiopharmaceutical relative to the normal database. Questions have been raised whether image reconstruction parameters affect the expressed Z-scores, potentially impacting the diagnostic accuracy. This study investigates how the use of the Bayesian Penalized Likelihood (BPL) reconstruction algorithm in CortexID, with various beta-values, and the effects of Anatomical Tracers for Longitudinal Assessment and Synthesis (ATLAS) and Zero-echo-time (ZTE) attenuation correction, influence the interpretation and quantification of brain PET/MR images.

Methods List-mode PET data from 25 patients who underwent 2-deoxy-2-[18F]fluoro-D-glucose (FDG) 3D static brain exams using a PET/MRI Signa 3T scanner (GE Healthcare) were reconstructed with the BPL algorithm at 𝛽-values of 𝛽50, 100, 200, 400, 600, 1000, and 1500. Attenuation correction (AC) methods, ATLAS AC and ZTE AC, remained constant as 𝛽-values varied, resulting in 14 images per patient. CortexID analyzed these images, obtaining Z-Scores. The study assessed Z-Score variations with different 𝛽-values and AC choices. 𝑠𝑦𝑛𝑔𝑜.via displayed images from one patient to visually assess parameter impact on quality and interpretation. Each image underwent analysis with two Volume-of-interests (VOIs): one on a uniform background and one on a region with high tracer uptake. Extracted max-, mean-, and peak-SUV values provided quantitative data.

Results In brain PET/MR images, higher 𝛽-values led to lower Z-Scores, and ATLAS AC showed slightly lower Z-Scores than ZTE AC. Lower 𝛽-values resulted in sharper but noisier images, while higher 𝛽-values yielded blurrier but less noisy ones. ATLAS AC produced slightly sharper images compared to ZTE AC, with an almost imperceptible difference. Max-, mean-, and peak-SUV decreased with increasing 𝛽-values in high tracer uptake regions. Mean-SUV in the uniform background remained nearly constant. Optimal image quality to the naked eye was achieved at 𝛽100-𝛽200.

Conclusion In CortexID, higher 𝛽-values in BPL reconstruction for FDG brain PET/MR images lead to decreased Z-Scores. ZTE AC shows higher Z-Scores compared to ATLAS AC, with minimal differences in image comparison. Higher 𝛽-values lead to a reduction in max-, mean-, and peak-SUV. Additional research is needed to assess the optimal 𝛽-value in CortexID image quantification.