Bayesian penalized likelihood PET reconstruction impact on quantitative metrics in diffuse large B-cell lymphoma

Area of Science:

• Nuclear Medicine
• Radiology
• Oncology

Background:

• Standard ordered subset expectation maximization (OSEM) and Bayesian penalized likelihood (BPL) are PET reconstruction methods.
• Diffuse large B-cell lymphoma (DLBCL) management relies on F-18 fluorodeoxyglucose PET/CT imaging.
• Reconstruction methods can impact quantitative and subjective image interpretation.

Purpose of the Study:

• To compare quantitative, subjective (Deauville score), and reader agreement differences between OSEM and BPL PET reconstruction in DLBCL.
• To assess the impact of reconstruction method on lesion detection and interpretation.

Main Methods:

• Retrospective review of 104 F-18 FDG PET/CT exams from 52 DLBCL patients.
• Comparison of OSEM and BPL reconstructions by an unblinded radiologist moderator and four blinded radiologists.
• Evaluation of quantitative metrics (SUVmax, SUVmean) and subjective Deauville scores (DS).

Main Results:

• BPL reconstruction showed significantly higher SUVmax and SUVmean compared to OSEM (P < .001).
• Deauville scores were altered in 25% of cases when comparing BPL and OSEM by the same reader.
• Interobserver agreement for DS was higher with OSEM (lesions >1cm: 0.89; ≤1cm: 0.84) than BPL (lesions >1cm: 0.85; ≤1cm: 0.81).
• Intraobserver agreement between OSEM and BPL was moderate (lesions >1cm: 0.67; ≤1cm: 0.4).
• Volumetric PET measurements (e.g., metabolic tumor volume) were similar between BPL and OSEM.

Conclusions:

• BPL PET/CT reconstruction in DLBCL results in higher SUV values, altered Deauville scores, and different reader agreement compared to OSEM.
• Harmonization of PET data, potentially through accreditation, is crucial to balance sensitivity and consistency in lesion detection.
• Standardization efforts are needed to ensure reliable interpretation of PET imaging across different reconstruction techniques.