MRM Highlights
All Posts
By Agah Karakuzu
Welcome to the first Editor’s Pick of 2019! We are starting off this year’s Q&A series with a reproducibility assessment study by researchers from the University of Zurich. We interviewed Yoojin Lee and Zoltan Nagy about their paper on establishing intra- and inter-vendor reproducibility of T1 relaxation time measurements at 3T, which shines a light onto one of the less prominent features of variable flip angle (VFA) T1 mapping.
MRMH: Can you tell us a bit about yourselves and how you came to be involved in MRI research?
Yoojin: I studied electrical engineering for my bachelor’s, but I have always been interested in how the human body works. So, MRI was the ideal option for me. I saw my current position advertised on the ISMRM website and applied. That’s how I started working with Zoltan.
Zoltan: I have a very similar story. I studied physics and math, but I, too, was always interested in the brain. My master’s was already in physiology with single-cell patch-clamp recordings of synaptic transmission. I attended a summer program at the Karolinska Institute in Stockholm, and afterwards I switched to a PhD in MRI. It was exciting to go from studying single cells to the entire brain.
MRMH: Can you give us a brief summary of your paper and explain how this work fits in with your broader research goals?
Yoojin: More broadly, our aim is cortical parcellation, especially using diffusion datasets. We extract fingerprints (we call them feature vectors) and use them to distinguish cortical areas. To introduce more information into this feature vector, we decided to acquire quantitative T1 maps. First, however, in seeking to establish that the datasets we use are comparable to those of others, we discovered that there was actually quite a big difference – about 10%. The difference could originate from several sources. For example, differences in B1 mapping methods implemented by vendors. We examined many possible sources and detailed them in our paper. We also performed a longitudinal study, where we scanned one subject in two Siemens scanners of the same model, and still found differences of around 1-2% which was higher than the test-retest variability. In addition, another subject was scanned on the same Philips scanner before and after a software upgrade. Interestingly, the software upgrade led to a 2-4% difference in T1 values.
MRMH: You did everything in your control to minimize vendor-based differences in data acquisition and used highly conservative ROI masks for the analyses. Yet, even so, inter-vendor variability was notable. Insights? Is there any way to adjust for it?
Yoojin: MRI scanners are built to aid humans in making diagnoses and in effect the scanners themselves are not required to be quantitative measuring devices. We for example expect a thermometer to give us a measurement that is pretty close to the true temperature. But the actual numbers in MR images can be anything as long as a human eye can detect an injury/disease/abnormality. In the actual pixel intensities there are so many sources of variability that it is impossible to match them across different scanners. Such differences make multicenter studies really difficult. I would say that manufacturers should work more on standardizing certain acquisition methods, such as B1 mapping.
Zoltan: The research community has come a long way toward quantitative imaging, but given that MRI scanners are such complex instruments, we need the involvement of the vendors. This involvement could amount to as little as more convenient access to the scanner calibration and sequence parameters. Better yet, co-supervision of PhD students in an industry/academia setting would be great. Of course the above-mentioned collaborations may not be possible for logistic, IP or financial reasons, so I’ll be happy if a quantitative MRI scanner just arrives on the market.
MRMH: If multiple imaging centers around the world were to decide to make their VFA data for T1 mapping publicly available, what suggestions would you have for them?
Yoojin: Because of the complexity of MRI scanners and the sensitivity of T1 relaxation time measurements to the design, settings or state of the individual components of a scanner, we would need as much information as possible, especially details about the RF pulse they use, the spoiler gradient moments and the RF phase cycling factor. For example, we know from experience, and mentioned it in our accompanying YouTube video, that in B1+ mapping with the actual flip angle method, setting the spoiler gradients properly can increase test-retest reproducibility. But really it would be far better if they were to provide the pulse sequence itself.
MRMH:The potential use of relaxation time mapping in clinics is a decades-old debate. What is your take on it?
Zoltan: I don’t doubt that quantitative imaging could provide useful biological markers for clinical practice. But even if the methods for quantitative MR imaging were firmly nailed down the results would only be as good as the state of the scanner. At our research sites quality assurance scans are run regularly (weekly or even more often than that). In contrast, some clinics hardly ever do quality assurance scans and, depending on the site, the scanners may be older and hence unable to cope with all the current state-of-the-art methods. In some cases, vendors come in a few times in a year and do their routine adjustments. If something breaks down, you might not know about it for months. As such, “a clinical method” may not mean the same thing at every site. I always say that when we buy a scanner it is rather like a Toyota, mass produced on a conveyor belt. But the thing is, we need Formula 1 performances from it! Research is therefore very much about “pimping” our scanners.
Yoojin: I think it will take some time for relaxation time mapping to be used in clinics. If you want to use it in diagnostics, then the effect size of the specific disease should be significantly higher than the T1 variability we measured, which was sizable in the multicenter setting. We should solve this issue first, then maybe we can use T1 maps in clinics.
Zoltan: AT the advanced neuroimaging workshop organised by ISMRM in Korea several presentations and the ensuing discussions dealt with reproducibility. We agreed that granting agencies are not very interested in funding reproducibility studies. In my view, if you have $100 million to allocate to research, it would be beneficial for the community as well as the agency to allocate a certain percentage of that allowance to quality assurance and reproducibility research. Otherwise we/they risk wasting resources on efforts that produce false positives or negatives simply because nobody ever bothered or had a chance to check the variance components of the chosen method. Having said this, we cannot deny that there is an increased awareness of the importance of reproducibility. The quantitative imaging biomarkers alliance (QIBA) has been working toward this goal for years, the OHBM Replication Award will be given out for the third time this year in Rome, and the Quantitative MR and Reproducible Research ISMRM study groups were recently formed as well. We are happy to be part of this movement and delighted that our efforts and results were selected as the Editor’s Pick.
