Assessing Adequacy and Variability in Semi-Structured Ga-68 PSMA PET/CT Reports for Prostate Cancer in view of auto-input to AI models
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Introduction
Gallium-68 Prostate specific membrane antigen (Ga-68 PSMA) PET/CT has significantly improved prostate cancer imaging by offering superior sensitivity and specificity over conventional modalities. However, the effectiveness of this diagnostic tool depends on the quality and consistency of reporting. This study evaluates the adequacy, consistency, and AI compatibility of semi-structured Ga-68 PSMA PET/CT reports.
Methods
Essential reporting elements were determined through consensus among nuclear medicine physicians, urologists, and radiotherapists. Two hundred Ga-68 PSMA PET/CT reports from prostate cancer patients (January 2020–June 2024) were analysed. Reporting Adequacy Score (RAS) assessed the percentage of clinical needs met, while Variability Index (VI) quantified inconsistencies in terminology. Statistical analyses, including descriptive statistics, frequency distribution, and boxplots, were performed to evaluate reporting trends.
Results
RAS ranged from 31% to 78%, with 91% of reports being partially adequate (50– 80% RAS) and 9% inadequate (<50% RAS). Key clinical details, such as Lesion Maximum Standardized Uptake Value (SUVmax), Neurovascular bundle involvement, and Bone metastasis lesion count, were frequently missing. Inconsistent terminology was observed in lesion descriptions, lymph node involvement, and uptake patterns, with VI ranging from minimal (4%) to high (67%). Reports with lower RAS and high VI were less suitable for AI- based data extraction, posing challenges to automation.
Conclusion
The majority of Ga-68 PSMA PET/CT reports were partially adequate, with significant missing details and considerable variability in terminology. The variability was evenly distributed across minimal, moderate, and high levels. Training an AI model on such reports would likely result in slower learning and compromised performance.
