My understanding of an MRI is that it takes a cross-section image of your body and pieces it back together digitally. So it’s not exactly taking an instant snapshot, but instead is compositing several images taken over a duration of time.
Now, if your heart is pumping during the scan (which I’m hoping it does), then wouldn’t the size of the heart vary between each image?
Do they do something to account for this in the software, or do they physically do something during the scan?
In: 35
I asked my supervisor this when I was at uni and she said one method is to record the heartbeat and synchronise the image capture so that it takes the scans at the same point of each pump 🙂
I have had both a regular and cardiac MRI. The regular one was no big deal, just kind of hang out and not focus on the confined space. The cardiac one was exhausting. Pretty much a 2 hour crunch/pilates/yoga/cardio session in a hamster tube.
My sense is that they took some backing images, but all of the important imaging was some variation of hold your breath and tense your muscles for 20 seconds, or exhale and tense your muscles for 20 seconds. Basically, intentionally ensure that you are removing any consciously controlled movement so that they can focus just on the heart and try to snap images throughout the beat.
Like taking screen shots of a moving video. 98% will be crap, but you will get some clear images. Now make 50 attempts over 2 hours, and each attempt requires you to hold a 20 second crunch while remaining perfectly still, in a tube. I literally had to rest in my car for 15 minutes before I drove home. Physically beat up.
I’ve worked on this a little. It’s usually a combination of gating and superresolution.
**Superresolution** — MRI scanners get faster at lower resolutions. You can do a very low res or small volume scan in less than 100 milliseconds, which is fast enough to freeze most heart movement.
You can’t see much in these very low res images, so instead you take 100s of them, with slight movement of the scanner each time, and then reconstruct a high-res image from all the low res images. The image reconstruction techniques in modern video games work a little like this.
**Gating** — you attach a heart monitor and note the exact point in the heart cycle of each tiny scan you make. When you want to reconstruct the final movie, you put all the tiny scans into maybe 32 buckets, with one bucket for each 50ms period of the cycle, and then do the superresolution reconstruction on just the scans in that bucket. Put the 32 final images together and bingo, you have a movie you can loop.
People have experimented with extra techniques, like estimating motion vectors to remove movement and increase sharpness, but I don’t know much about that. No doubt you could use ML to help as well.
tldr: the movies you see are composites of many, many heart cycles recorded over a long period of time.
Latest Answers