Are You Normal? A Case on the Impact of Reconstruction Algorithms on PET Neuroimaging Reference Data

Introduction - Accurate interpretation of neuroimaging scans relies heavily on the establishment and utilization of reference data or a "normal database”. Therefore, understanding the methodologies behind the creation of these reference datasets is paramount. This case study shows the importance of adhering to reference data and investigates the impact of reconstruction algorithms on their reliability.

Materials and methods- The patient data were initially analysed using a novel software package employing an in-house developed reference database. These results were compared against our established standard reference dataset. To delve deeper into the nuances of the impact of the databases, comparisons were then conducted using a variety of reconstruction algorithms.

Results - Our findings noted considerable discrepancies in the interpretation of neuroimaging scans when different databases and reconstruction algorithms were applied. During visual interpretation of the images, the case exhibited characteristics indicative of a positive neurodegenerative scan, which was then evaluated semi-quantitatively. Using the established database with the occipital lobe as normalisation region, the obtained z-score for the Left-/Right Putamen were [-2.28, -3.39], confirming the visual neurodegenerative scan. Conversely, using the established reconstruction with the new database, the same region yielded z-scores of [-1.19, -1.9], potentially contradicting the visual diagnosis. Using the same reconstruction as was used in the new database, z-scores of [-1.92, -2.38] were obtained, leading to a different clinical diagnosis. As did the analysis with the new reconstruction algorithm, matched for original iterations, with the new database, obtaining z-scores of [-1.69, -2.27, again potentially contradicting the visual positive diagnosis.

Conclusion - As professionals in nuclear medicine, it is imperative to question the assumptions underlying our reference data. This case study emphasizes the impact of such reference data on clinical diagnoses, urging the community to reassess and refine the methodologies employed in establishing and using "normal" benchmarks for neuroimaging. By enhancing our understanding of the intricacies involved in creating reference data, we can elevate the accuracy and reliability of clinical interpretations, ultimately improving patient outcomes.