Meeting with Meike Lerner of European Hospital, Dr Persson spoke of ‘synthetic MRI’, new software that enables the radiologist to do a single MR scan without changing scanner settings – and more. ‘A lot of MRI control parameters on the scanner – such as repetition time TR, the echo time TE or the application of pre-pulses, can be transferred to the PACS. Usually, a technician sets the scanner parameters and the radiologist has to accept the images he receives. The new software means you can change TR and TE at will, generating any T1 or T2-weighted image, after the actual scan is performed. Even a FLAIR image (FLuid Attenuated Inversion Recovery) or fat suppression can be generated as a post-processing step. All these contrast controls, which are usually on the scanner, are now transferred to the radiologist, so he can check his own best optimal contrast settings. This may save a lot of time, because he will have all his preferred images based on a single scan.’
Showing a whole-head 25-slice scan, he pointed out that it was obtained in just five minutes. The software enables this and the radiologist can use his own personal optimization. ‘If he wants to look at T2 weighted images in this way, he gets it this way. It is easy to teach the staff, and you also have only one data set to send,’ he pointed out. ‘MR imaging usually takes from 30 minutes to an hour. I can’t claim I can do an hour in five minutes, but you can probably save quite some time — probably 30%. So, for every three patients now, you could have an extra patient.
The advantage of the technique is that it is based on quantitative values. Each tissue has a characteristic set of MR parameters such as T1 and T2 relaxation and Proton Density (PD). If part of an image contains these values you can directly classify the tissue. On the whole head image he indicated ‘healthy’ reference numbers for white matter, grey matter and cerebro-spinal fluid (CSF). ‘If your measured values correspond to these reference values you can be considered healthy. It may be that the image pixels contain more than one single tissue (partial volume); in that case the measured values should lie on a line in between these reference values. If you are outside these healthy areas you are probably looking at a pathology. Multiple Sclerosis lesions, for example, have completely different values for T1, T2 and PD than any normal brain tissue
Could MS automatically be recognised? ‘It is not implemented yet, but you can see that it is more or less easy to calculate the percentage white matter and grey matter, and you can actually get out partial volume and show an image that is tissue specific: an image that contains only white matter or an image that contains only grey matter on a scale of 0-100%. The next step is then, obviously, to take the whole brain and then remove the grey and the white matter and the CSF, then you would only have the disease left. If you are healthy it would be a black image. If not, those areas would show up brightly.’
A volume estimation is also obtainable, he added. ‘It is a simple subtraction of the volume of the whole brain minus the healthy areas. What is left is the volume of the pathology. So instead of trying to grey scale the contrast images to find MS lesions, you get real answers that are independent of thresholding or grey scaling.’ So the radiologist can make a safer diagnosis? ‘You see an absolute match — it’s not your eyes or how you set the image to look at. You can get an automatic indication of where the computer found values that are not normal. Of course the radiologist is still needed to decide whether the software was right or not, but it may save him a lot of time. Especially important is also that if a patient has a uniform change throughout the brain, you cannot know, with a contrast image, what you are looking at. You have the same change everywhere, and so it looks normal again. Using synthetic MRI, however, this change will be found. I suspect if you have a MS patient with normal appearing white matter this technique is better than the conventional method
Could early-stage Alzheimer’s be detected in this way? ‘We don’t know yet. Hopefully, yes. We have started to apply this method on various patients and the results are not yet analysed. Anyhow, here you can see the benefits of MR quantification. You can see in this little window down here that all the tissue – CSF, grey matter and white matter — always end up in the same place. So each tissue has its own characteristic values, the same combination of T1, T2 and PD. Hence it does not float like a normal contrast image. It is just a value that is fixed.’
But what if a radiologist needs to use contrast agents? ‘Yes, you could do a scan before and after contrast. That’s a possibility. But maybe the scanning is so good that you don’t need any contrasting anymore, because you have all the contrast images. Or you could do it before the actual scan. You only take a quantified image after the contrast medium injection, which saves more time.’
Couldn’t this destroy the whole contrast agents industry?
‘Yes,’ he replied.
Adaptability for various MRI scanners
In principle, Dr Persson explained, if a scanner has rapid quantification, the software will be compatible. ‘The whole thing is based on quantification. The three MR parameters are T1 and T2 relaxation and proton density PD. Nobody does quantification that well – it is there, but hardly applied because other quantification methods take so much time. That is why everyone reverts to contrast imaging. You make a contrast image and that is the image you get, you can’t change it anymore. The whole concept here is that you do this single quantification scan, you know the T1, T2, and PD, then you can calculate how an image would look with certain scanner settings. In principle, it is precisely like a Hounsfield unit on CT, it is really a quantified measure, but in our case we have three outfield units simultaneously, we have T1, T2 and PD.’
Asked what would happen if 1.5T or 3T were in use, he pointed out that more signal would be gained on 3-Tesla: ‘That’s why it can be even faster. This is a five minute scan for 1.5 Tesla. On a 3-T scanner, it would only take two minutes, after which you have all of the T1- and T2 weighted images.’
During the two-year development of this software a specific implementation was made for cardiac examinations, for the so-called Late Gadolinium Enhancement. This has already been in use for a more than a year. Based on a single scan of 20 seconds LGE images can be generated that cover the complete heart and show any desired inversion delay. Here again, the optimal contrast can be set after the actual scan, full control for the user. ‘All the clinicians insist this is fantastic. They said: We want this. Because they have control; they light up like, Wow, of course! It’s really nice. People come from other universities to see it.
Sectra customers will have the option to receive this new software as a plug in. ‘Of course, the software needs to match the hospital’s particular model. But theoretically there is no problem with any scanner.’
This is not simply new software — it is a completely new tool, he pointed out. Apart from its clinical validation for quantification, progress will now come from learning further uses. ‘We are doing a lot of autopsies on corpses. That is our real gold standard for heart examinations. We can do the histology to compare our results. We can save a lot of time that way. It’s also good for research, because we can develop and test the software. There is a lot of validation. At this year’s ISMRM in Toronto there will be several presentations regarding this new system from CMIV. Although the system is not yet marketed, many people now work with it. Everyone has jumped at this.’
CMIV is running seven projects involving autopsy. As autopsy is forbidden in many countries; additionally many families often do not want their deceased to be touched, there is worldwide demand for this, Dr Persson, who is working to develop virtual autopsy, using this new system. ‘We can do this single quantification scan of the person and the study results can be produced an hour later.’ This has aroused enormous interest, including the making of a Discovery Channel programme, and visits to the university by all international TV companies. Following a visit to Washington, to present the software to the Bush administration (described as bringing autopsies to life) this research has gained a big grant. ‘There’s worldwide demand for this,’ said Dr Persson, ‘and it’s coming.’
‘But imagine if you have your CT and someone says to you: We are going to skip the Hounsfield unit; we are going to make it into a floating scale that can be anything. You would be outraged. But really, using MRI it can be anything you are looking at; the images have no absolute value. A lot of people tried to do this, it was really popular in the ‘80s, but it failed because nothing was fast enough. We have found a method that is fast enough.
‘There’s already a lot here, but I can’t wait to get more,’ he continued. ‘We want to use it on the spine, the knee and so on. Look at this cardiac sequence from hospital imaging. The healthy myocardium should be black. That means that you have to set your scanner precisely so that it becomes black, which is very difficult; it depends on a patient’s weight, how much contrast media you actually put in it, how long you waited between contrast injection and the scan, and so on. So a lot of people struggle to get their inversion time, this parameter, right. With synthetic MRI you can automatically get the optimal inversion time; I just point at the area in the image that I want to be black and the software automatically recalculates the corresponding optimal inversion time that makes it black. Look, I take this little square and I place it somewhere that I want to be black — this is black blood, this is black myocardium. You can see there is a big pathology here, lighting up. This means that, based on a single scan, you always have the optimal result, and you don’t have to test your way forward, which is currently the usual case. That is about four or five scans. In our institute we tracked the reduction of examination time and on average it was 7 minutes 45 per patient. And that directly saves money. This is only a very simple application for the heart – and the patient only holds his breath one time. It is also safer, particularly for MRI for children, because their parameters change – particularly if very young. The newborn look completely different than a one year old. It is really hard to optimize your sequences. But with this system, you can do it afterwards, with the brain sequence, for example, and get your optimum settings. And instead of ten different specs, you only have to look at one. It’s easier to compare the old study with the new study. So it becomes faster and safer for the patient; you can never forget an image. With the absolute numbers it is easier to compare a patient over time or to send the data to another hospital with completely different software and completely different scanners, and have the same results.’
In clinical use, radiologists have seen patients’ examinations double in numbers, which of course, he pointed out, has led them to say they have more work to do. ‘But we can also help them a lot by pre-processing. We provide all these numbers and warning images for diseases. If it says zero, all the images were black, for example. If it says 20 percent then you know there’s something there. We can even try to code clustering and then try to assign some sort of illness before you actually see the image. There are a lot of possibilities. This is the future. I think we will integrate plug ins, because you should have all of the possibilities to change everything on the PACS viewer, but not change your scanner.’
Sectra’s next-generation PACS
Dr Persson divides his time 50-50 between medical and technical research. He is very involved in the development of Sectra’s next-generation PACS, IDS7, a web-based PACS system that presents a new way to handle very large data sets. Thanks to a fresh approach in image content and transfer, radiologists can review images in a matter of seconds. This whole solution is built with the data explosion in mind. Even when images eventually reach terabyte-size, only relevant information will be transmitted.
Before this, these large datasets could only be viewed on modality workstations. With Sectra’s patent-pending technology transferring the volumes to the usual PACS workstation is no longer an issue. . ‘Now, you don’t need to “shake hands” with every slice that comes over the network. You do it once, then the whole volume loads. If you want to see the skeleton, you can only pick that up from the archive. It saves a lot of time. We have worked with this for a very long time. The system is being used by several customers, including Telemedicine Clinic in Barcelona and Södertälje Hospital in Sweden.’
* Initiated by Linköpings University, Landstinget i Östergötland and Sectra AB, the multidisciplinary Centre for Medical Image Science and Visualisation (CMIV) conducts research to provide methods and tools to meet future clinical needs in image analysis and visualisation. (http://www.cmiv.liu.se/)