TL;DR
This study highlights that variations in data processing for 18F-FET PET scans, even when using the same scanner, can lead to significant differences in key quantitative metrics used to characterize brain gliomas. The authors emphasize the urgent need for standardization to ensure comparability of results across different research and clinical centers.
What Problem Does This Paper Address?
Positron Emission Tomography (PET) using the amino acid O-(2-[18F]fluoroethyl)-L-tyrosine (18F-FET) is a valuable tool for diagnosing and monitoring brain tumors. It helps differentiate tumor tissue from normal brain, assess tumor grade, and track treatment response. However, the interpretation of 18F-FET PET data relies on quantitative parameters like tumor-to-brain ratios (TBR) and tumor volume (Tvol). The problem is that different research and clinical centers may employ slightly different methods for processing these PET scans. This variability can lead to inconsistent results, making it difficult to compare findings across studies or even between different departments within the same institution. The authors aim to quantify the impact of these methodological differences on key 18F-FET PET parameters.
Methods
The study compared data processing methods from two German centers (Center A and Center B) that used the same PET scanner but had distinct protocols. They analyzed 18F-FET PET data from 40 patients with untreated cerebral gliomas. The key methodological differences investigated were:
- Framing of dynamic PET data: How the time-series data collected after tracer injection was divided into time bins.
- Data reconstruction: The algorithms used to create the 3D PET images from raw data (filtered back projection (FBP) vs. iterative reconstruction (ITR)).
- Tumor delineation cut-off: The threshold value (TBR) used to define the boundaries of the tumor for volume and uptake measurements.
- Region of Interest (ROI) definition for time-activity curves (TACs): How the specific areas within the tumor were selected for analyzing tracer uptake over time.
They then compared parameters like TBRmax (maximum tumor-to-brain ratio), TBRmean (mean tumor-to-brain ratio), Tvol (tumor volume), time-to-peak (TTP), and the slope of the TAC. Receiver Operating Characteristic (ROC) analysis was used to assess the diagnostic performance of these parameters in differentiating high-grade from low-grade gliomas.
Datasets
The study utilized a dataset of 40 adult patients with previously untreated cerebral gliomas (20 WHO grade II, 10 WHO grade III, 10 WHO grade IV). Histopathological diagnoses were available for all patients. The data was collected between June 2006 and April 2014 at the Forschungszentrum Jülich. No new datasets were introduced; the study focused on re-analyzing existing data using different processing pipelines.
Key Insights and Ideas
- Processing matters, even with the same scanner: The study powerfully demonstrates that even when using identical PET hardware, variations in software-based data processing can lead to significantly different quantitative results.
- TBRmax and Tvol are particularly sensitive: These two key metrics showed the most substantial differences between the two processing methods, suggesting they are highly susceptible to variations in reconstruction and delineation.
- Standardization is crucial for clinical translation: The findings underscore that without standardized protocols, it's challenging to establish reliable cut-off values and compare results across different centers, hindering the widespread clinical adoption and validation of 18F-FET PET.
- Tumor grading accuracy remains robust: Despite differences in absolute parameter values, the ability to differentiate between high-grade and low-grade gliomas using ROC analysis remained comparable between the two methods, suggesting that the relative differences might be more important than absolute values for this specific task.
Key Results
- TBRmax: Center B consistently reported higher TBRmax values (3.34 ± 1.13) compared to Center A (2.84 ± 0.99), with a statistically significant difference (p < 0.001).
- Tvol: Center B also reported larger tumor volumes (1.51 ± 1.44 ml) than Center A (1.14 ± 1.28 ml), again with a significant difference (p < 0.001).
- TTP: Dynamic analysis showed a longer time-to-peak in Center B (30.8 ± 6.3 min) compared to Center A (22.4 ± 8.3 min), a significant difference (p < 0.001).
- TBRmean: While there was a statistically significant difference (p < 0.04), the mean values were very close (Center A: 2.20 ± 0.41 vs. Center B: 2.16 ± 0.41), and this difference might not be significant with multiple testing correction.
- Slope: A significant difference was observed in the slope of the TAC (Center A: -0.06 ± 0.92 SUV/h vs. Center B: 0.73 ± 0.69 SUV/h, p < 0.001).
- ROC Analysis: The accuracy in differentiating high-grade from low-grade gliomas (measured by AUC) was similar for both centers across various parameters (TBRmax, TBRmean, Tvol, TTP, slope), with no statistically significant differences.
Limitations and Future Work
- Limited sample size for ROC: The authors note that the biopsy-controlled study used to inform one of the center's thresholds had a relatively small sample size (52 biopsies in 31 patients), which might not be sufficient to establish a universal clinical standard.
- WHO Classification: The study used the 2007 WHO classification for gliomas, which predates the inclusion of molecular parameters in the 2016 classification. This could affect the direct applicability of the AUC values to current classifications.
- Reconstruction method impact: The study strongly suggests that the use of filtered back projection (FBP) in Center A, compared to iterative reconstruction (ITR) in Center B, significantly contributed to the observed differences, particularly in TBRmax and Tvol. Newer PET systems increasingly use iterative reconstruction, which may reduce these discrepancies in the future.
- Future Work: The authors explicitly call for standardization of data processing and protocols for 18F-FET PET to ensure comparability of clinical results. Further studies are needed to establish optimal thresholds, especially for pre-treated patients, and to incorporate molecular data into glioma classification.
Summary updated