The effect of SUV discretization in quantitative FDG-PET Radiomics: the need for standardized methodology in tumor texture analysis.
Leijenaar RT, Nalbantov G, Carvalho S, van Elmpt WJ, Troost EG, Boellaard R, Aerts HJ, Gillies RJ, Lambin P. · 2015
Read The effect of SUV discretization in quantitative FDG-PET Radiomics: the need for standardized methodology in tumor texture analysis. by Leijenaar RT, Nalbantov G, Carvalho S, van Elmpt WJ, Troost EG, Boellaard R, Aerts HJ, Gillies RJ, Lambin P. (2015) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc4525145.txt.
Introduction
Ralph T. H. Leijenaar leads a nine-author team warning about a hidden setting in cancer scan analysis, in Scientific Reports in 2015. Radiomics reads patterns in tumor scans to gauge how uneven a tumor is, a possible marker of behavior. To compute texture features from PET scans, brightness values must first be grouped into bins, and labs choose how in unlike ways.
The team tested two binning rules on scans of 35 lung cancer patients, taken before and during treatment. Forty-four texture features changed value with the bin width used. One rule, holding bin width fixed, kept values comparable across patients and visits; the other, fixing the bin count, did not. Patients even swapped ranks between rules, so the authors call for one shared standard.
Catalog record
| CREATOR | Leijenaar RT, Nalbantov G, Carvalho S, van Elmpt WJ, Troost EG, Boellaard R, Aerts HJ, Gillies RJ, Lambin P. |
|---|---|
| YEAR | 2015 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 35 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
The effect of SUV discretization in quantitative FDG-PET Radiomics: the need for standardized methodology in tumor texture analysis By Ralph TH Leijenaar, Georgi Nalbantov, Sara Carvalho, Wouter JC van Elmpt, Esther GC Troost, Ronald Boellaard, Hugo JWL Aerts, Robert J Gillies, Philippe Lambin Published in: Scientific Reports DOI: 10.1038/srep11075 ABSTRACT Abstract FDG-PET-derived textural features describing intra-tumor heterogeneity are increasingly investigated as imaging biomarkers. As part of the process of quantifying heterogeneity, image intensities (SUVs) are typically resampled into a reduced number of discrete bins. We focused on the implications of the manner in which this discretization is implemented. Two methods were evaluated: (1) RD, dividing the SUV range into D equally spaced bins, where the intensity resolution (i.e. bin size) varies per image; and (2) RB, maintaining a constant intensity resolution B. Clinical feasibility was assessed on 35 lung cancer patients, imaged before and in the second week of radiotherapy. Forty-four textural features were determined for different D and B for both imaging time points. Feature values depended on the intensity resolution and out of both assessed methods, RB was shown to allow for a meaningful inter- and intra-patient comparison of feature values. Overall, patients ranked differently according to feature values–which was used as a surrogate for textural feature interpretation–between both discretization methods. Our study shows that the manner of SUV discretization has a crucial effect on the resulting textural features and the interpretation thereof, emphasizing the importance of standardized methodology in tumor texture analysis. In recent years, oncological research has increasingly focused on the prediction of treatment outcome based on individual patient and tumor characteristics1, aiming to avoid the one-size-fits-all treatment approach that under- and over-treats a large number of patients. Imaging can play a crucial role here, as it allows for a non-invasive identification and characterization of the tumor2,3. Positron emission tomography (PET) is a valuable tool for detecting and staging cancer4. In recent years, PET imaging has also been increasingly used for decision support5, treatment planning6,7 and response monitoring during radiotherapy8. The most widely used PET tracer is [18F] fluoro-2-deoxy-D-glucose (FDG), commonly quantified by standardized uptake values (SUVs)9. Easily derived SUV measurements, such as the maximum or peak SUV10, are described as predictors for treatment outcome11,12,13,39. Additionally, more advanced quantitative imaging features describing tumor image texture (i.e. the spatial arrangement of intensities within the image), which reflect intra-tumor heterogeneity of metabolic activity, are increasingly being investigated as potential imaging biomarkers in lung14,15, head and neck16,17, cervical16,18, esophageal19,20,21 and other cancers22,23 — a field of research often referred to as ‘Radiomics’2,3,24,25,40,41,42. Efforts have been made to provide guidelines for quality control measures in PET imaging and to standardize patient preparation, dose administration, image acquisition, image reconstruction and SUV normalization, in such a way that absolute SUV measurements are interchangeable in multicenter studies26. Interchangeable SUV measurements are very important in PET Radiomics, but the methodology used to determine textural features is also subject to variability. Standardization is therefore needed27,28,29 (Fig. 1). [Figure: Levels of standardization in PET Radiomics.Interchangeable absolute SUV measurements are obtained by standardizing patient preparation, dose administration, image acquisition, image reconstruction and SUV normalization26. Standardization of the methodology used for tumor texture analysis ensures interchangeable PET Radiomic features and their ascribed values] One important methodological factor is SUV discretization (i.e. resampling image intensity values). Discretization reduces the otherwise infinite possible number of intensity values to a finite set and effectively reduces image noise. Most recent literature describes using a fixed number (e.g. 8, 16) of discrete resampled values or ‘bins’ to divide the image SUV range into equally spaced intervals before calculating textural features14,15,16,18,19,20,21,22,28,30,31,32. Consequently, this results in discretized images with varying bin sizes or ‘intensity resolutions,’ depending on the SUV range. An alternative discretization method is to resample the image SUVs with a fixed bin size in units of SUV (e.g. 0.1, 0.5), maintaining a constant intensity resolution across all tumor images33. When aiming to identify imaging biomarkers in cohort and multicenter studies or trials, it is important that textural features and their ascribed values be directly comparable, both inter- and intra-patient, in order to derive meaningful conclusions. To our knowledge, the effect of the SUV discretization method in this respect has not been previously evaluated and we hypothesize that the aforementioned intensity resolution used for SUV discretization plays a key role in this regard. The general objectives of our study are to compare both aforementioned conceptually different discretization methods for several popular textural features and to identify which of these methods is most appropriate for texture quantification in a clinical setting. We will specifically investigate the role of the intensity resolution and use a clinical case study to demonstrate the effect of the SUV discretization methodology on the interpretation of the assessed textural features. MATERIALS AND METHODS Patients and PET imaging This study comprised 35 non-small cell lung cancer (NSCLC) patients who were prospectively included in a clinical trial (NCT00522639) and scheduled for radiotherapy and/or chemotherapy between July and December 200811. 18F-FDG-PET/CT imaging was performed on a Biograph 40 PET/CT scanner (Siemens Medical Solutions) twice: (1) after induction chemotherapy but before radiotherapy and (2) during the second week of radiotherapy (Fig. 2a,b). Patients fasted for at least six hours before imaging. The injected amount of 18F-FDG was (4 × body weight [kg] + 20) MBq. Patients rested 60 minutes before image acquisition. Patients’ blood glucose levels were below 10 mmol/L, so no correction for blood glucose level was applied. [Figure: Left column: Representative images of sequential imaging for one patient, showing pre-treatment imaging (a) and imaging during the second week of radiotherapy (b). The tumor delineation is outlined in green. Both images are displayed with the same window/level settings. Right column: Histograms of the pre-treatment and during treatment images, resampled with a fixed bin size (i.e. intensity resolution) (c) or a predefined number of bins (d). In (d), one can appreciate the difference in resulting intensity resolution when resampling with a fixed number of bins. Pre-treatment and during treatment intensity resolutions were 0.6 and 0.37 [SUV], respectively] PET images were iteratively reconstructed using normalization- and attenuation-weighted OSEM using 4 iterations, 8 subsets and a 5 mm Gaussian filter. The resulting images had an in-plane pixel size of 4 × 4 mm and a 3 mm slice thickness. PET images were converted into units SUV, normalized by patient body weight9. Tumor volumes of interest (VOIs) were manually delineated on fused PET/CT images for treatment planning purposes. Further details are described elsewhere11. This study was conducted according to national laws and guidelines and approved by the appropriate local trial committee at Maastricht University Medical Center (MUMC+), Maastricht, The Netherlands. All included patients signed an informed consent form. Image processing and feature extraction SUVs within the VOI were first discretized using: (1) a fixed bin size (B), or intensity resolution, in units of SUV (Fig. 2c) and (2) a fixed number of bins (D), or discrete resampling values (Fig. 2d). For image I, let I(x) represent the SUV of voxel x, SUVmin the minimum SUV in I and SUVmax the maximum SUV in I. Resampling SUVs into bins with an intensity resolution of B was performed using: Where term ensures that the bin count starts at 1. We