[01:08.650 --> 01:09.310]  All right.
[01:09.490 --> 01:09.990]  Hello, everyone.
[01:10.150 --> 01:11.430]  Welcome to the next talk today.
[01:11.670 --> 01:13.030]  I hope you're having a good day so far.
[01:13.350 --> 01:14.330]  I've seen some interesting talks.
[01:15.270 --> 01:15.690]  All right.
[01:15.730 --> 01:16.450]  Just a couple of things.
[01:16.650 --> 01:17.570]  Remember to stay high rated.
[01:18.150 --> 01:20.050]  Be sure to keep your mask on.
[01:20.090 --> 01:21.690]  We appreciate you wearing them all the time.
[01:21.850 --> 01:23.530]  It makes it much easier for all of us.
[01:23.790 --> 01:27.390]  And, you know, thank you from the conference itself for that.
[01:28.330 --> 01:30.230]  We have a fourth track.
[01:30.430 --> 01:32.210]  You've probably seen it floating around somewhere.
[01:32.350 --> 01:36.870]  If you're interested in giving a talk or presentation and you aren't scheduled for one, you can go to the info desk and you can ask.
[01:36.870 --> 01:41.610]  And there's some slots in the coffee shop coffee house where you can go and give, you know, 15 or 20 minute presentation.
[01:41.930 --> 01:44.970]  So after seeing all these wonderful presentations, you feel inspired.
[01:45.270 --> 01:46.470]  Head down there and sign up.
[01:46.610 --> 01:47.670]  We'd love to hear what you have to say.
[01:49.010 --> 01:51.390]  There is hackers got talent tonight.
[01:51.590 --> 01:56.490]  So hopefully some of you have some great talent and you're willing to go sign up and participate in that.
[01:56.730 --> 01:58.590]  So hope to hope to see you there.
[01:59.070 --> 02:02.910]  So today's talk is about MRI machines and how they work.
[02:02.910 --> 02:05.010]  And our presenter is Doug Brantner.
[02:05.110 --> 02:07.950]  And with that, Doug is presenting via Zoom.
[02:08.230 --> 02:09.790]  So you won't see his face today.
[02:09.790 --> 02:10.670]  You'll see his talk.
[02:11.050 --> 02:13.170]  If you have questions, we're going to have a Q&A period.
[02:13.290 --> 02:20.670]  You can either submit them to the Matrix chat for this channel or you can let me know and I will read them in because he can't hear you in the audience.
[02:20.930 --> 02:23.910]  So with that, with no further ado, off to Doug.
[02:25.130 --> 02:26.910]  Hi, my name is Doug Brantner.
[02:27.290 --> 02:28.970]  And this is How Does an MRI Machine Work?
[02:31.610 --> 02:32.850]  So just a quick warning.
[02:33.030 --> 02:34.370]  We're going to see some medical images.
[02:34.510 --> 02:37.630]  We're going to talk about health and diseases and talk about hospital and medical equipment.
[02:37.830 --> 02:38.890]  But nothing too graphic.
[02:39.050 --> 02:40.210]  It should be safe for all ages.
[02:40.850 --> 02:42.710]  I might talk fast because I have a lot to cover.
[02:43.670 --> 02:46.690]  I have no personal financial interest to disclose related to this talk.
[02:46.690 --> 02:49.170]  That's something they make you say before medical conferences.
[02:49.790 --> 02:53.750]  And the views in this talk do not represent those of my employer or the open source imaging project.
[02:54.350 --> 02:55.850]  Most of this talk is disclaimers.
[02:56.310 --> 02:57.350]  I am not a doctor.
[02:57.550 --> 03:00.830]  I'm not an MRI technician or an MRI safety expert or a medical physicist.
[03:01.350 --> 03:05.970]  If you have any questions related to health or interventions, I cannot answer those.
[03:06.150 --> 03:11.070]  So you should seek a licensed doctor or a medical technician or all of the above.
[03:12.070 --> 03:15.190]  Because I've only been doing this for two years and I'm not there yet.
[03:15.190 --> 03:17.930]  So I am an MRI research engineer.
[03:18.570 --> 03:23.030]  I work on sensors for motion correction and motion phantoms.
[03:23.250 --> 03:27.690]  Phantom is any test object that is not a person that you put in the MRI for testing.
[03:28.230 --> 03:31.530]  And I also work on simulations and body models.
[03:31.870 --> 03:35.750]  I have a master's in scientific computing, which is half math and computer science.
[03:36.550 --> 03:39.950]  I like to say I have a master's degree in floating point errors.
[03:40.430 --> 03:42.350]  I did my thesis in computer vision.
[03:42.350 --> 03:48.250]  This is solving stereo distances from stereo images and a little bit of machine learning.
[03:49.110 --> 03:50.490]  I went back to school.
[03:50.710 --> 03:51.730]  I used to work on movies.
[03:51.970 --> 03:53.210]  This is where I used to work.
[03:53.950 --> 03:55.090]  This is my office.
[03:55.370 --> 03:56.910]  This is the view for my office.
[03:57.110 --> 03:58.210]  I was in this one.
[03:59.030 --> 04:02.730]  And this is the first science lab that I ever worked on, if anyone recognizes that.
[04:03.950 --> 04:05.950]  And I've been a HOPE volunteer since 2010.
[04:06.230 --> 04:08.310]  Shout out to the AV crew and the streaming crew.
[04:09.250 --> 04:10.630]  Sorry, I can't be there this year.
[04:12.770 --> 04:14.730]  So a little bit quick thank yous.
[04:14.970 --> 04:16.230]  Thanks to HOPE in 2600.
[04:16.590 --> 04:20.810]  There was a talk at my first HOPE in 2010 on hacking the education system.
[04:20.810 --> 04:23.730]  And that kind of put the bug in my head to go back to school.
[04:24.590 --> 04:28.430]  And none of this would have happened without the open source community and the hacker community.
[04:29.250 --> 04:34.850]  And a double thank you to the Adafruit jobs board for finding my last two jobs for me.
[04:36.810 --> 04:38.510]  So I'm just going to leave these up here.
[04:38.690 --> 04:42.210]  I don't think I have enough time to go into them, but a little bit on back to school.
[04:42.210 --> 04:44.130]  You can go back and pause the video if you're interested.
[04:50.870 --> 04:55.910]  So I work on motion sensors for MRI because motion and MRI is a huge problem.
[04:55.910 --> 05:00.510]  And it causes all kinds of artifacts in the images or just messes up the images.
[05:01.170 --> 05:08.130]  Anything from bulk motion of your body to respiratory motion from breathing or even your heartbeat can mess it up.
[05:08.890 --> 05:12.670]  And basically, there's not much you can do about it except for repeating the scan.
[05:12.670 --> 05:14.930]  And MRI is pretty slow, so it takes a lot of time.
[05:15.090 --> 05:16.910]  So that is not an ideal solution.
[05:17.610 --> 05:19.770]  There's a lot of effort going into motion correction.
[05:20.490 --> 05:25.430]  And the current methods, there's like one sensor where they put a pneumatic belt around you.
[05:25.790 --> 05:29.310]  And it's supposed to help with what they call respiratory gating.
[05:29.730 --> 05:32.590]  But it doesn't work very well and it's very cumbersome.
[05:32.690 --> 05:34.190]  So we're looking for better solutions.
[05:35.130 --> 05:39.910]  So this is a poster we submitted to an MRI conference this year.
[05:40.470 --> 05:44.410]  This is a 3D USB camera that can take videos.
[05:44.670 --> 05:50.450]  And we basically put it in some copper pipe with some conductive glass for shielding and stuck it in the MRI.
[05:50.450 --> 05:52.570]  And we were able to get a respiratory motion.
[05:53.170 --> 05:55.770]  This you have to be very careful because of the magnet.
[05:56.250 --> 05:58.730]  There's a lot of safety testing that happens before this.
[05:58.870 --> 06:00.350]  So don't try this at home.
[06:00.970 --> 06:08.890]  And one thing that's kind of interesting is that the MRI, the magnetic field increases very quickly as you approach the MRI.
[06:09.510 --> 06:16.570]  And if you move any conductive object, obviously you can't use any metal or magnetic objects.
[06:16.570 --> 06:18.550]  So we use copper because it's not magnetic.
[06:19.790 --> 06:26.530]  But if you move any conductor, it induces a current and you can burn things out just by moving around in the MRI room.
[06:26.530 --> 06:31.170]  So one of the cameras actually burned out just from moving it into the MRI too quickly.
[06:31.390 --> 06:32.310]  So you have to be really careful.
[06:33.190 --> 06:37.110]  Another cool thing about MRI, these are headphones for hearing protection.
[06:37.110 --> 06:38.370]  And you'll see there's an air tube.
[06:38.950 --> 06:45.190]  So they pipe the sound in using an air tube because you want to avoid conductors in the MRI if at all possible.
[06:45.210 --> 06:48.430]  Because they can heat up and cause burns if they're not handled properly.
[06:49.470 --> 06:52.790]  So we want to show that our sensors don't interfere with the MRI.
[06:53.410 --> 06:57.370]  So SNR or signal to noise ratio is a huge factor in MRI.
[06:58.210 --> 07:02.330]  And we want to make sure that we're not negatively affecting the SNR of the scanner.
[07:02.490 --> 07:05.790]  And we show before and after without the camera and with the camera.
[07:05.930 --> 07:06.510]  And it's pretty comparable.
[07:07.010 --> 07:10.650]  And the MRI signal is entirely radio frequency.
[07:11.230 --> 07:14.870]  So we want to make sure that we're not causing any RF interference in the scanner too.
[07:14.870 --> 07:15.910]  So this is the baseline.
[07:16.150 --> 07:17.830]  It has a funny shape, but that's normal.
[07:18.250 --> 07:21.930]  And this is with the camera operating.
[07:21.930 --> 07:26.470]  And you see a spike in the noise frequency in one or two frequencies.
[07:27.230 --> 07:30.650]  But typically, the MRI only cares about the center.
[07:30.850 --> 07:36.830]  So these may or may not be tolerable depending on the type of imaging you're doing.
[07:37.370 --> 07:39.750]  And we could definitely improve the shielding as well.
[07:40.570 --> 07:42.350]  So this is the respiratory waveform.
[07:42.350 --> 07:44.530]  I think this is a video.
[07:44.810 --> 07:45.070]  Yeah.
[07:45.750 --> 07:48.230]  So this is a video of someone in the scanner breathing.
[07:48.270 --> 07:50.410]  And you can see we can track the motion quite nicely.
[07:51.390 --> 07:52.430]  It's also kind of interesting.
[07:52.470 --> 07:53.790]  You can see the buttons in their shirt.
[07:55.610 --> 07:58.690]  The buttons actually go the wrong way though, probably due to diffraction.
[07:58.890 --> 08:00.570]  So that's something that we should look into further.
[08:01.030 --> 08:03.670]  But this is a pretty promising for motion tracking.
[08:03.670 --> 08:08.410]  And this, you know, we could either feed the signal into the scanner and do real-time motion correction.
[08:08.450 --> 08:12.570]  Or you can do retrospective motion correction in post-processing.
[08:14.450 --> 08:18.270]  So last year, the year before, this is my colleague's work.
[08:18.890 --> 08:20.210]  We use an accelerometer.
[08:20.470 --> 08:22.090]  This might look familiar to some people.
[08:22.270 --> 08:24.830]  We use an accelerometer, which measures acceleration.
[08:25.150 --> 08:26.990]  And we use that as a proxy for motion.
[08:27.470 --> 08:30.090]  And then we use the wireless motion microcontroller.
[08:30.550 --> 08:32.790]  And we found a non-magnetic battery.
[08:33.390 --> 08:38.730]  So we could transmit real-time motion signals to a Raspberry Pi over Wi-Fi.
[08:38.890 --> 08:40.870]  So this one's also wireless, which is kind of cool.
[08:41.030 --> 08:43.970]  And that's really important, again, because of the heating.
[08:44.930 --> 08:48.810]  And wires can cause heating and cause burns if you're not careful.
[08:48.950 --> 08:50.270]  So wireless is really ideal.
[08:51.090 --> 08:53.550]  And we have to do a lot of magnetic safety testing.
[08:53.550 --> 08:56.870]  So we have to test each part individually before we even go in the room with it.
[08:57.510 --> 08:58.910]  To make sure it's not magnetic.
[08:59.150 --> 09:04.250]  So even though this has this big metal chunk here, it seems to be not magnetic.
[09:04.470 --> 09:06.910]  So we were able to use it in the scanner safely.
[09:07.210 --> 09:09.430]  And we also have to do heating testing with the RF.
[09:09.590 --> 09:11.650]  And make sure that nothing's going to heat up and cause burns.
[09:12.990 --> 09:15.910]  So we also want to make sure that we're not going to mess up the images.
[09:16.430 --> 09:21.250]  So this is a B0 map or a map of the magnetic field that the scanner can take itself.
[09:21.730 --> 09:23.130]  And this one is pretty clean.
[09:23.270 --> 09:24.710]  So you want it to be as even as possible.
[09:24.710 --> 09:26.850]  And this is a water bottle phantom.
[09:27.870 --> 09:29.830]  Because most MRI imaging is water.
[09:31.010 --> 09:33.710]  And we put the different test objects on top of it.
[09:33.750 --> 09:36.450]  And we see how much they mess up the magnetic field.
[09:36.590 --> 09:43.070]  So even if they're not going to fly into the scanner, they can still disrupt the homogeneity of the magnetic field.
[09:43.510 --> 09:44.990]  So this one is pretty clean.
[09:44.990 --> 09:46.790]  These are two different accelerometers.
[09:47.250 --> 09:51.010]  And this one causes a disturbance in the magnetic field.
[09:51.230 --> 09:56.030]  So it might have, you know, maybe the coating on the PCB might have like nickel or something in it.
[09:56.110 --> 09:59.910]  So it's not enough to fly into the magnet, but it is enough to disturb the magnetic field.
[10:00.470 --> 10:01.990]  And here we show two different microcontrollers.
[10:02.350 --> 10:10.230]  And they both, you know, probably because of this big metal piece, this one takes a pretty big chunk out of the magnetic field and we can't even get an image there.
[10:10.750 --> 10:14.910]  So I think we wound up, I forget which one we used, but probably this one.
[10:14.910 --> 10:16.670]  And then two different batteries.
[10:16.950 --> 10:22.230]  Again, they're both not magnetic, but this one, you know, really messes up the magnetic field and this one doesn't.
[10:22.630 --> 10:24.350]  And again, we look at the RF noise too.
[10:25.790 --> 10:34.190]  So in order to test them, we want to make sure that the MRI is not interfering with our sensors and we're actually getting good signals that are actually useful.
[10:34.930 --> 10:46.270]  So I worked on a motion phantom, which is basically a stepper motor controller using an Arduino that drove a cam to squeeze an airbag.
[10:46.690 --> 10:48.390]  And then we send an air tube through.
[10:49.150 --> 10:51.470]  So this is the MRI is in a Faraday cage.
[10:51.830 --> 10:59.830]  And we have to be very careful passing anything through what we call a wave guide, which is basically a pipe that goes through the cage.
[10:59.830 --> 11:03.950]  And this is to block RF noise from the outside world from interfering with the MRI.
[11:04.530 --> 11:07.670]  So we pass an air tube, which is totally safe.
[11:07.830 --> 11:11.010]  And so everything in the MRI room is rubber from the motion phantom.
[11:11.390 --> 11:19.170]  And then we put the sensor box on top of it and it transmits the motion data over Wi-Fi to a Raspberry Pi that just barely fit in there.
[11:19.610 --> 11:27.210]  And since the Wi-Fi signal is so far out of our MRI frequency range that it doesn't really affect the image.
[11:27.210 --> 11:29.510]  But we might have gotten some low harmonics, which is interesting.
[11:30.090 --> 11:37.150]  So the motor is programmed with a pre-described waveform and we can compare that against the sensor data.
[11:37.610 --> 11:42.790]  But then our goal is to measure motion and patience and we can't program the patience.
[11:43.230 --> 11:46.670]  So my idea, I guess, having a film background was to put a camera.
[11:46.910 --> 11:50.230]  So we use the camera for tracking objects that we can't program.
[11:51.530 --> 11:53.390]  And this is a video of that.
[11:53.850 --> 11:57.750]  So this is an in vivo scan, meaning a real-life person, a volunteer.
[11:58.790 --> 12:00.350]  We have to do a lot of safety testing.
[12:00.510 --> 12:05.750]  The sensor is in a shielded box made of copper and we have to do heating testing to make sure it's not going to cause burns.
[12:05.990 --> 12:08.990]  And we put it inside a flame-proof pouch anyway, just to be safe.
[12:09.230 --> 12:14.910]  And you could see this is comparing the accelerometer signal in blue to the video signal in orange.
[12:16.350 --> 12:19.990]  So we also tried putting the stepper motor in the MRI room.
[12:22.550 --> 12:24.090]  Do not try this at home.
[12:24.710 --> 12:28.250]  And so it causes an enormous amount of radio frequency noise.
[12:28.510 --> 12:32.010]  So we put it in a shielded box that was custom-made by our machinist.
[12:33.450 --> 12:36.030]  It has a sliding door so we can put the motor in and out.
[12:36.130 --> 12:39.530]  It has a shielded DB9 connector, which we use for the motor power.
[12:39.950 --> 12:43.850]  We wind up replacing a lot of screws with brass screws and stuff like that.
[12:46.050 --> 12:52.790]  And it has a waveguide again, which is an open conduit for the mechanical output.
[12:53.050 --> 12:56.310]  So we could put like a shaft or a belt or a string through it.
[12:58.490 --> 13:00.850]  So we can actually get the motion out of the box.
[13:02.350 --> 13:04.170]  So yeah, do not try this at home.
[13:04.330 --> 13:06.690]  This means it's not safe to use in the MRI room.
[13:06.690 --> 13:08.790]  So do as I say, not as I do.
[13:10.050 --> 13:12.530]  But the magnet is actively shielded.
[13:12.690 --> 13:15.250]  So the magnetic field actually falls off pretty quickly.
[13:15.970 --> 13:20.630]  So all the way back here, there's almost no magnetic pull on the box.
[13:20.750 --> 13:26.650]  But just to be safe, we tied some strings to it so that it wouldn't slide because the scanner vibrates while it's on.
[13:27.470 --> 13:30.750]  So this is like, yeah, do not try this at home.
[13:32.430 --> 13:37.830]  And so this is the baseline RF noise again, radio frequency.
[13:38.630 --> 13:43.190]  And then this is what happens with the old cable, which wasn't shielded very well.
[13:43.250 --> 13:45.410]  And it really blows up the noise really badly.
[13:45.770 --> 13:47.970]  So we would not get a good picture from the signal.
[13:48.590 --> 13:53.510]  And then with the new cable, with a double shielded twisted pair cable, the twisted pair really helps.
[13:54.090 --> 13:59.810]  And messing around with the grounding, we got it back down to baseline, which I was like really proud of.
[14:01.290 --> 14:04.510]  And just to make sure it's working, we opened that sliding door.
[14:04.730 --> 14:06.090]  And this is what happened.
[14:06.330 --> 14:08.430]  So I think that shows that the shield was actually working.
[14:10.450 --> 14:13.890]  So this is a picture of the motion phantom close-up.
[14:14.030 --> 14:15.550]  So the stepper motor driving a cam.
[14:15.830 --> 14:18.510]  The airbag and the air tube goes into the MRI room.
[14:18.650 --> 14:20.150]  And this is the sensor sitting on top of it.
[14:20.610 --> 14:22.130]  This is a head coil.
[14:22.390 --> 14:26.130]  So the MRI signal is radio frequency again.
[14:26.590 --> 14:29.830]  And so we need antennas in the MRI to pick it up.
[14:29.990 --> 14:34.690]  And so this is actually full of antennas that would pick up the signal from a head scan.
[14:35.430 --> 14:38.390]  And I don't know if you can see, but this is the water bottle phantom inside.
[14:41.230 --> 14:45.050]  And here's the video of the airbag working and the sensor on top of it.
[14:45.130 --> 14:48.250]  And this is the signal coming from the sensor over Wi-Fi.
[14:48.830 --> 14:50.050]  This is a Wi-Fi antenna.
[14:54.900 --> 14:59.240]  And then these are the signals we presented at the conference.
[14:59.400 --> 15:01.640]  This is the same from the video.
[15:02.820 --> 15:07.320]  One thing... the first thing I ever did was I put some LEGOs in the MRI, which was kind of cool.
[15:08.000 --> 15:09.860]  So I got paid to play with LEGOs.
[15:10.160 --> 15:11.800]  The motor cannot go in the MRI.
[15:11.920 --> 15:12.660]  That would not be safe.
[15:12.660 --> 15:16.100]  But, you know, wood, brass screws, those are safe.
[15:16.220 --> 15:17.580]  You have to be careful with the cables again.
[15:18.260 --> 15:21.660]  This was a test jig to test a linear encoder.
[15:23.840 --> 15:29.620]  And I used the rubber bands instead of springs and then a string that we could pull that went through the waveguide.
[15:29.620 --> 15:33.660]  So everything is non-conductive except for the cables.
[15:34.660 --> 15:38.400]  So MRI, it uses magnets and radio waves.
[15:38.640 --> 15:41.720]  There is no ionizing radiation in MRI.
[15:42.100 --> 15:47.400]  So x-rays and CT or CAT scans use radiation and MRI does not.
[15:47.560 --> 15:48.840]  So that's one of the major benefits.
[15:50.620 --> 15:55.820]  We're mostly looking at your water molecules, which are full of hydrogen protons.
[15:57.120 --> 15:58.320]  The CAT scan...
[15:58.320 --> 16:03.660]  A CAT scan is actually spinning around you really fast.
[16:03.760 --> 16:08.820]  And it literally spins the x-ray around you really fast to get a 3D picture.
[16:09.280 --> 16:13.760]  And the MRI instead uses changing magnetic fields to get multiple views.
[16:13.880 --> 16:15.380]  And that's how we get 3D views.
[16:15.660 --> 16:18.640]  And in this case, this is from those multiple radio coils.
[16:18.820 --> 16:21.080]  So they're each sensitive to a different area.
[16:21.080 --> 16:26.440]  And then we combine them at the end to get a nice picture and what's called parallel imaging.
[16:27.460 --> 16:29.040]  These are a few other things.
[16:29.180 --> 16:30.100]  So this is...
[16:30.100 --> 16:32.340]  You can take an MRI picture at any angle.
[16:33.020 --> 16:35.480]  And they're all cross sections of the inside of the body.
[16:35.600 --> 16:38.020]  So this is an oblique slice or a diagonal slice of the heart.
[16:38.020 --> 16:39.920]  So you can see all four chambers.
[16:40.380 --> 16:43.720]  This is a knee in the sagittal view or sideways view.
[16:44.220 --> 16:49.560]  And this is an axial view, which is looking in the head foot direction or down the bore of the scanner.
[16:50.160 --> 16:52.160]  And this is like the liver and the spleen.
[16:53.480 --> 16:55.340]  And these are some other images.
[16:55.600 --> 16:58.380]  Here you can see a meniscal tear.
[16:58.740 --> 17:00.180]  These are T2 images.
[17:00.220 --> 17:01.800]  And the previous ones were T1.
[17:01.800 --> 17:02.840]  And we'll get to all that later.
[17:03.360 --> 17:04.380]  And then these are...
[17:04.380 --> 17:04.760]  You can...
[17:04.760 --> 17:11.180]  By manipulating the settings on the scanner, like you would with like a manual camera, you can get different...
[17:11.180 --> 17:13.580]  You can highlight different tissues or suppress different tissues.
[17:13.740 --> 17:16.780]  So in this case, on the left, this is a fat suppressed image.
[17:16.780 --> 17:19.040]  And on the right is a regular image.
[17:19.180 --> 17:25.380]  And you can see the subdermal fat, but also the bone marrow gets darker when you turn on the fat suppression.
[17:27.180 --> 17:28.180]  This is...
[17:32.750 --> 17:33.930]  I hope you can hear this.
[17:34.870 --> 17:36.450]  I'm going to scroll through this actually.
[17:36.970 --> 17:37.630]  Oh, I'm sorry.
[17:39.010 --> 17:41.350]  This was partially a video project.
[17:42.170 --> 17:48.490]  So basically, you can scroll through the slices and get a 3D view of the body.
[17:48.490 --> 17:51.070]  And this is like how the radiologist would look at it.
[17:51.370 --> 17:55.610]  And you could see like the liver and the spleen and the kidneys and the spine.
[17:55.770 --> 17:58.050]  And you can see the major blood vessels, which is cool.
[17:59.090 --> 18:01.450]  And then you can take the slices in any direction.
[18:04.050 --> 18:04.470]  Yeah.
[18:04.810 --> 18:12.530]  So you can kind of think of it as a 3D volume, but the pictures are taken in one dominant direction.
[18:12.530 --> 18:14.490]  So one direction is always going to look the best.
[18:15.070 --> 18:16.370]  And you can scroll...
[18:16.370 --> 18:19.590]  You can also turn them into 3D models, which is really interesting.
[18:19.790 --> 18:25.290]  And we actually use these for simulations from MRI images.
[18:25.310 --> 18:30.030]  And this is also a big topic in dataset anonymization and patient privacy.
[18:30.410 --> 18:33.290]  Because if you can do this, you could do it with a face.
[18:33.470 --> 18:43.610]  So you can like scrub all the metadata, but you know, there's actually a lot of research going on in what's called defacing MRI images to like completely anonymize the dataset in today's world.
[18:48.460 --> 18:48.940]  Yeah.
[18:49.100 --> 18:50.320]  And you can scroll through them.
[18:51.900 --> 18:53.820]  So you can select the slices.
[18:53.980 --> 18:55.480]  You can select the orientation of the slices.
[18:55.640 --> 18:58.360]  So this is a diagonal slices taken in the brain.
[18:58.500 --> 19:01.320]  So this is 40 slices and they're taken one at a time.
[19:01.640 --> 19:02.720]  And each time...
[19:02.720 --> 19:05.120]  Each slice takes time to acquire.
[19:05.580 --> 19:07.720]  And that's why motion is such a big problem.
[19:07.720 --> 19:08.840]  So if you...
[19:08.840 --> 19:09.460]  If one...
[19:09.460 --> 19:10.420]  Each...
[19:10.420 --> 19:14.520]  If you take one slice and then you move your head a little bit, then it's gonna...
[19:14.520 --> 19:17.380]  The next slice is not gonna line up or you're gonna get blurring or something like that.
[19:18.540 --> 19:20.560]  This is some COVID-19 lung imaging.
[19:20.820 --> 19:23.480]  You can see some damage in the lungs.
[19:24.020 --> 19:27.720]  And the bottom is a comparison with the CAT scan.
[19:28.120 --> 19:32.240]  And you can see the MRI has a lot nicer tissue contrast in the soft tissues.
[19:32.680 --> 19:35.780]  And it doesn't expose you to the ionizing radiation.
[19:35.780 --> 19:38.460]  So that's a huge benefit of a MRI.
[19:39.820 --> 19:41.560]  F MRI is kind of interesting.
[19:41.860 --> 19:45.440]  They use oxygen level as a proxy for metabolism.
[19:46.020 --> 19:47.100]  So the...
[19:47.100 --> 19:48.500]  You know, are they reading your thoughts?
[19:48.680 --> 19:50.400]  You know, there's a lot of research on that right now.
[19:50.960 --> 19:58.940]  And they can also do functional MRI of other body parts, like the lungs, using different concentrations of oxygen.
[19:58.940 --> 20:01.020]  And on top is healthy lungs.
[20:01.060 --> 20:04.440]  And you can see the oxygen perfusion is pretty even throughout the lung.
[20:04.600 --> 20:06.940]  And on the bottom is people with diseased lungs.
[20:07.140 --> 20:08.040]  And you can see the difference.
[20:10.480 --> 20:12.940]  Interventional MRI is anything that...
[20:13.820 --> 20:16.060]  Any procedure that's guided by an MRI.
[20:16.260 --> 20:20.380]  So they might put you in MRI and then, you know, use it for a biopsy or a surgery.
[20:20.380 --> 20:26.340]  Or they use MRI to guide radiation therapy to avoid damaging nearby tissues.
[20:26.960 --> 20:31.400]  Low field MRI or lower magnet strength is a very hot new topic.
[20:31.700 --> 20:37.100]  Because the magnet is a lower strength, it opens up a lot of possibilities
[20:41.750 --> 20:45.260]  for stuff you can bring into the MRI room safely.
[20:46.680 --> 20:48.080]  So MRI is very slow.
[20:48.080 --> 20:50.300]  It has a very limited field of view.
[20:50.440 --> 20:52.540]  So you could get a full body CT in less than a minute.
[20:52.920 --> 20:58.620]  Where a full body MRI would probably take, you know, maybe an hour and it would be at least six partial scans.
[20:59.000 --> 21:00.300]  Very sensitive to motion.
[21:00.540 --> 21:04.160]  Any implants or metal in your body is a hazard, potentially.
[21:05.100 --> 21:11.800]  Some more modern implants are MR conditional, depending on the scanner and the settings.
[21:11.800 --> 21:21.180]  Some tattoos and makeup pigments actually contain ferromagnetic particles, which can heat up in the MRI.
[21:21.320 --> 21:22.300]  So you have to be careful with that.
[21:24.040 --> 21:25.820]  Patient comfort is not ideal.
[21:26.140 --> 21:30.740]  A lot of people have claustrophobia and the MRI is not a great place for that.
[21:31.020 --> 21:34.500]  I'm kind of hoping with this talk, maybe we can help alleviate some of that.
[21:34.600 --> 21:40.180]  And maybe you'll be interested in all the technical stuff that's happening around you and help with that a little bit.
[21:41.320 --> 21:42.280]  And you can't move.
[21:42.400 --> 21:43.760]  You have to stay perfectly still.
[21:44.500 --> 21:48.340]  And you might even have to hold your breath for some scans, which not everybody can do.
[21:48.720 --> 21:50.240]  And finally, it's very expensive.
[21:50.260 --> 21:52.360]  It's a huge, complex installation.
[21:52.360 --> 21:54.320]  You kind of have to build the building around the MRI.
[21:55.300 --> 21:58.940]  And because of that, it has very low worldwide access.
[21:59.300 --> 22:05.080]  I've heard anecdotally that there's more MRIs in New York City than there are in some countries.
[22:06.080 --> 22:08.200]  And so it can be very difficult to access.
[22:08.440 --> 22:11.660]  And that's part of what the Open Source Imaging Project is trying to solve.
[22:12.700 --> 22:18.560]  So quickly, your body is mostly water, which contains trillions of protons and the hydrogen atoms.
[22:18.740 --> 22:21.040]  And each proton acts as a tiny little magnet.
[22:21.240 --> 22:22.780]  And they're all randomly arranged.
[22:23.280 --> 22:28.920]  So when we put you inside the magnet, the magnetic field is extending along the long axis of the bore.
[22:29.620 --> 22:32.020]  And it causes all the protons to align.
[22:32.380 --> 22:34.500]  And it also causes them to precess.
[22:34.860 --> 22:40.420]  And that is like a gyroscope spinning around the Earth's gravitational field.
[22:40.560 --> 22:46.180]  So when you apply an external force, the magnets start to wobble like a gyroscope.
[22:47.740 --> 22:51.940]  And that can emit radio signals because they're tiny little spinning charges.
[22:53.780 --> 22:55.200]  But they're all out of phase.
[22:55.200 --> 22:56.880]  So they're not doing anything useful yet.
[22:57.400 --> 23:03.780]  So what we do is we apply a radio frequency pulse at 90 degrees to the scanner axis.
[23:05.740 --> 23:08.800]  And it tips all the protons 90 degrees.
[23:08.920 --> 23:11.020]  And it also causes them to line up in phase.
[23:11.200 --> 23:12.680]  So now the signal starts adding up.
[23:13.280 --> 23:16.580]  And then we listen to the signal with the RF coils.
[23:16.920 --> 23:19.100]  And that's the MRI signal.
[23:24.280 --> 23:26.460]  But we still don't have any spatial information.
[23:26.480 --> 23:28.460]  We don't know where the signal is coming from.
[23:28.460 --> 23:34.200]  So we can do what's called NMR or nuclear magnetic resonance spectroscopy.
[23:34.580 --> 23:40.860]  And we can look at the chemical composition of the signal by looking at the frequency spectrum of the signal.
[23:41.340 --> 23:44.880]  But we don't have enough information to make an image yet.
[23:45.140 --> 23:46.700]  Because we have no spatial information.
[23:46.700 --> 23:53.720]  And nuclear magnetic resonance refers to the nucleus of the hydrogen proton.
[23:53.720 --> 23:56.440]  So it's not anything radioactive.
[23:56.840 --> 24:00.160]  And they dropped the word nuclear for obvious reasons.
[24:00.320 --> 24:03.040]  And especially to differentiate between CT and CAT scans.
[24:03.040 --> 24:05.320]  Because there is no radiation in MRI.
[24:06.700 --> 24:12.400]  So in order to get the spatial information, we add more coils, which are called gradient coils.
[24:12.580 --> 24:14.560]  And a gradient just means a slope.
[24:16.000 --> 24:23.480]  And what they do is they tip the magnetic field and cause the magnetic field to vary in a known way.
[24:23.480 --> 24:27.660]  And this affects the frequency that the protons will resonate at.
[24:27.880 --> 24:38.100]  And by doing that, we can cause a known frequency dependence on position because we're varying the spatial magnetic field in a known way.
[24:38.100 --> 24:40.940]  And that's how we get spatial information.
[24:41.820 --> 24:47.660]  And the gradients are very fast switching, really high current electromagnet coils.
[24:48.080 --> 24:51.940]  And because of the fast switching, they cause a mechanical vibration.
[24:51.940 --> 24:54.420]  And the scanner is kind of shaped like a speaker.
[24:54.440 --> 24:56.980]  And that's where the classic MRI sound comes from.
[24:58.060 --> 25:05.740]  And then finally, we acquire the images in what's called a Fourier space or a K space or frequency space.
[25:06.260 --> 25:14.120]  And so the final step is image reconstruction, where we apply a Fourier transform to convert the frequency space to an image.
[25:15.520 --> 25:16.540]  How are we doing on time?
[25:17.440 --> 25:18.780]  So this is some early MRI.
[25:19.040 --> 25:22.880]  I would be remiss because I know there's a lot of connections between HOPE and SUNY Stony Brook.
[25:23.060 --> 25:26.780]  So one of the early pioneers, Paul Lauterberg, won the Nobel Prize.
[25:27.340 --> 25:32.860]  And if this doesn't look like a hacker lab to you, I think he also deserves an honorary hacker prize.
[25:34.020 --> 25:42.920]  And then there was another one, another guy at Stony Brook, who developed one of the first superconducting magnets and did the first human MRI.
[25:43.520 --> 25:45.240]  And this looks terrifying.
[25:45.500 --> 25:47.760]  We don't wrap you in coils anymore like that.
[25:47.900 --> 25:52.220]  We might lay coils on top of you, but our coils, I promise, are much better insulated than that.
[25:52.820 --> 25:59.780]  The magnetic fields were, you know, very low field between 500 to 1000 Gauss, which is a measurement of the strength of the magnetic field.
[26:00.620 --> 26:04.280]  And this is the first human MRI in 1977.
[26:04.540 --> 26:09.880]  And you can see the lungs, the air doesn't make any signal, but you can kind of see the tissue surrounding the lungs.
[26:10.020 --> 26:11.200]  And this is a year later.
[26:11.380 --> 26:15.740]  And they tried to segment some of the organs, the stomach and the spleen and the lungs.
[26:16.100 --> 26:18.900]  And as you can see, we've come quite a long way since then.
[26:18.900 --> 26:31.980]  And this one took four and a half hours to make because they had to move the volunteer to acquire every single pixel in the image because it had a fixed magnetic field and it didn't have the gradients.
[26:32.180 --> 26:36.980]  So they actually had to move the subject throughout the field of view in order to get the image.
[26:38.220 --> 26:40.700]  So this guy might look familiar to some of you.
[26:41.300 --> 26:46.660]  So we measure modern MRIs in Tesla, which is equal to 10,000 Gauss.
[26:46.760 --> 26:47.680]  So they're quite a bit stronger.
[26:48.600 --> 26:51.160]  The average junkyard magnet is about one Tesla.
[26:51.400 --> 26:57.120]  And it's, you know, hundreds of times stronger than your refrigerator magnet and tens of thousands of times the Earth's magnetic field.
[26:57.580 --> 27:02.640]  Low field is generally less than one Tesla and high field is 7 to 12 Tesla for human imaging.
[27:05.400 --> 27:07.920]  The most common ones are one and a half or three Tesla.
[27:08.500 --> 27:12.160]  And then other research NMR magnets can go from 20 to 40 Tesla.
[27:12.400 --> 27:14.980]  And I think the most powerful is 45 Tesla.
[27:15.600 --> 27:17.060]  But those are not for humans.
[27:17.880 --> 27:19.700]  So the magnets, there's different shapes.
[27:19.840 --> 27:21.600]  These are what are called open MRIs.
[27:22.340 --> 27:24.780]  And they generally tend to be lower field strength.
[27:25.440 --> 27:29.380]  And they're not as common as the solenoid or cylindrical magnet.
[27:29.600 --> 27:31.060]  So this is what we're going to talk about today.
[27:31.560 --> 27:37.880]  And this is the main, the B0 axis or the main axis of the magnetic field is this arrow.
[27:38.120 --> 27:40.560]  So it's right down the center of the scanner or the bore.
[27:42.780 --> 27:47.340]  So this is a map of the magnetic field and how it falls off as you get away from the magnet.
[27:47.480 --> 27:50.200]  And it falls off pretty quickly because of active shielding.
[27:51.600 --> 27:54.100]  And you see the red and the yellow zones, but those don't matter.
[27:54.100 --> 27:59.400]  What matters is the five Gauss line or the 0.5 millitesla line, which is this dotted line out here.
[27:59.960 --> 28:01.900]  And that's what we call the danger zone.
[28:02.140 --> 28:07.380]  And you try to build the building so that the five Gauss line is contained within the room.
[28:07.500 --> 28:08.500]  But that doesn't always happen.
[28:08.780 --> 28:12.640]  And again, the B0 axis is pointing along the center of the MRI.
[28:14.580 --> 28:17.240]  The magnetic field extends in three dimensions.
[28:17.240 --> 28:19.300]  So it actually goes above and below the floor too.
[28:19.460 --> 28:20.360]  So you have to be really careful.
[28:20.740 --> 28:23.640]  And again, this is why the whole building has to be taken into account.
[28:25.720 --> 28:28.420]  This is an MRI guided surgical suite.
[28:28.940 --> 28:33.080]  And they actually painted the floor so that they mark the five Gauss line.
[28:33.480 --> 28:36.520]  And so this is the MRI danger zone and the white zone.
[28:36.660 --> 28:38.440]  And then the blue floor is where it's like safe.
[28:38.560 --> 28:41.740]  But I'm sure anybody that works in here has to be like very, very careful.
[28:42.040 --> 28:45.240]  And there's probably a lot of training that goes into this so they don't make mistakes.
[28:47.080 --> 28:53.760]  So the dangers from the magnet are the force or the missile effects due to the magnetic field.
[28:53.940 --> 28:59.120]  And that's why you can't bring any ferrous metal into the room because it'll just fly into the magnet.
[28:59.460 --> 29:00.760]  And then there's also a torque.
[29:00.920 --> 29:04.700]  So any elongated objects will try to align themselves with the magnetic field.
[29:05.100 --> 29:10.120]  And it can apply hundreds of times an object's weight in force.
[29:10.120 --> 29:11.620]  So it's pretty dangerous.
[29:12.420 --> 29:16.120]  There's also a force due to induced currents, which we talked about with the camera earlier.
[29:16.640 --> 29:27.140]  So even if it's not magnetic, any conductive material will, when you're moving it through the magnetic field, will induce a current which opposes, creates its own magnetic field which opposes the force.
[29:27.220 --> 29:29.320]  So it's a very weird feeling.
[29:30.660 --> 29:32.360]  So the magnet is always on.
[29:33.200 --> 29:35.040]  This is what happens if you're not careful.
[29:35.260 --> 29:36.400]  So don't do that.
[29:38.700 --> 29:42.180]  That's why they divide the building into restricted access zones.
[29:42.860 --> 29:45.020]  And I'm looking at all you pen-testers right now.
[29:45.560 --> 29:47.560]  Do these signs also apply to you?
[29:47.720 --> 29:54.000]  And if you don't believe that, I'm sure the magnet will be happy to confiscate your lockpicks for you in a maybe violent way.
[29:54.160 --> 29:55.640]  So, you know, respect the signs.
[29:56.740 --> 30:00.100]  This is a ultrasound cart that they use for some procedure.
[30:00.100 --> 30:01.180]  And they actually tied it.
[30:01.240 --> 30:03.460]  I don't know if you can see, there's a cable here tied to the wall.
[30:03.920 --> 30:05.640]  So it can't go past this line.
[30:05.800 --> 30:06.700]  So that's kind of interesting.
[30:08.380 --> 30:17.320]  The MRI is inside a Faraday cage because the RF signals that your body emits from your protons are very, very weak.
[30:17.340 --> 30:27.360]  So we try to block out as much external RF noise as possible, including special doors with RF seals and special glass with a mesh in it that's also conductive.
[30:28.840 --> 30:32.440]  And then, so how do we get wires in and out for the sensors?
[30:33.040 --> 30:34.520]  We either use a patch panel.
[30:34.880 --> 30:41.640]  And you have to pay a lot of attention to grounding and shielding, which my past life as a live sound engineer actually prepared me quite well for.
[30:42.140 --> 30:45.660]  Or you can use the waveguide and pass things through the waveguide because this is just a pipe.
[30:47.600 --> 30:51.160]  Ideally, we always use shielded cables because of the RF concerns.
[30:51.160 --> 30:59.860]  And ideally, you would avoid running them through the waveguide because the shield can act as an antenna and transmit RF noise from outside to inside the room.
[30:59.880 --> 31:00.680]  And that's a problem.
[31:01.000 --> 31:02.760]  This one, I believe, is fiber.
[31:02.980 --> 31:06.440]  And fiber is actually okay because it's not metallic at all.
[31:06.740 --> 31:14.480]  So a lot of commercially available MRI sensors use fiber, but that's very, very expensive.
[31:15.340 --> 31:18.020]  We also have to use cable traps because the shields can heat up.
[31:18.600 --> 31:27.920]  And so the cable trap helps block what's called eddy currents that build up from the RF pulse and block that from causing heating in the cable.
[31:28.020 --> 31:29.440]  But that can actually cause like serious burns.
[31:29.600 --> 31:31.480]  So that's something that we have to be really, really careful about.
[31:32.760 --> 31:33.940]  Don't be this guy.
[31:35.700 --> 31:37.740]  So the magnet is actually a superconductor.
[31:39.480 --> 31:42.200]  And superconducting means it has zero resistance.
[31:42.820 --> 31:53.500]  And they use fancy wire filaments that when you get them below a certain temperature, which is very, very low, they become superconducting and they lose their resistance.
[31:54.860 --> 32:03.280]  The reason the magnet is so big is it's actually a giant tank of liquid helium, which is held at 4 Kelvin, and it's cryogenically cooled.
[32:06.140 --> 32:09.340]  And this is to achieve the superconducting.
[32:10.620 --> 32:12.240]  So it's pretty complicated.
[32:12.400 --> 32:13.200]  It's called a cryostat.
[32:13.720 --> 32:16.780]  And it has very elaborate thermal shielding.
[32:17.240 --> 32:22.700]  This is an NMR magnet, which has a very small bore.
[32:22.840 --> 32:24.260]  It's just this tube at the top here.
[32:24.400 --> 32:26.840]  So this is for like measuring chemical samples and stuff.
[32:28.100 --> 32:31.960]  This one actually has an inner chamber with liquid nitrogen to keep it even colder.
[32:32.120 --> 32:33.460]  So this one's probably a pretty high field.
[32:33.880 --> 32:37.500]  But there's like many, many layers of thermal insulation to keep the magnet cold.
[32:38.920 --> 32:49.460]  So when you inject the current into the magnet, you inject the current once when you start the magnet, and then you walk away and there's no power source and the current just keeps flowing forever.
[32:49.720 --> 32:52.940]  So have we, you know, magically discovered perpetual motion?
[32:54.200 --> 32:54.920]  Not quite.
[32:55.140 --> 33:00.240]  So there is a very small resistance and the magnet does actually get weaker over time, but it's a very, very small amount.
[33:00.440 --> 33:03.080]  And over the lifespan of the magnet, it's pretty negligible.
[33:03.560 --> 33:07.680]  And there is energy going into the system, mostly with the chilling.
[33:07.820 --> 33:09.780]  You're pumping the liquid helium.
[33:10.540 --> 33:14.840]  So there's electrical energy and heat energy, and you also refill the liquid helium periodically.
[33:14.840 --> 33:16.700]  So sorry, there's no free lunch.
[33:17.680 --> 33:23.420]  The magnetic field has to be very, very even or what we call B0 uniformity, which is those maps I showed earlier.
[33:24.380 --> 33:29.600]  And because the frequency of the RF signal depends on the magnet, the strength of the magnet.
[33:30.360 --> 33:37.460]  So if we just use one big coil, we only get a very small area of homogeneity in the center, and that's not big enough for imaging.
[33:37.680 --> 33:42.660]  So by using multiple coils, we can expand that range into a much wider area.
[33:42.660 --> 33:46.200]  And so the magnet coils are pretty complicated designs.
[33:46.960 --> 33:56.420]  And then we actually add additional superconducting coils called shim coils that are active, and they're automatically adjusted by the computer before every scan.
[33:56.420 --> 34:04.880]  And that helps us get the homogeneity very even, because when you go into the scanner, your protons are interacting with the magnetic field.
[34:05.040 --> 34:10.020]  And so just your presence or any presence in the magnet will disrupt the magnetic field.
[34:10.020 --> 34:12.560]  So you have to do the active shimming before every scan.
[34:13.120 --> 34:17.560]  And there's another set of superconducting coils going in the opposite direction.
[34:17.900 --> 34:23.060]  And those are the active shielding coils, which help block the magnetic field from...
[34:23.060 --> 34:25.560]  or it helps contain the magnetic field to a smaller area.
[34:25.560 --> 34:30.740]  But it has the secondary effect that the magnetic field increases very quickly as you approach the magnet.
[34:30.900 --> 34:36.900]  So for what I do, we have to be extremely careful when we're testing new objects, because they could just fly out of your hand like nothing.
[34:38.040 --> 34:40.540]  So the magnetic field aligns all your spins.
[34:40.800 --> 34:42.720]  I think I have to go through this part pretty quickly.
[34:43.300 --> 34:46.180]  It aligns the spins of all your protons.
[34:46.660 --> 34:49.080]  And it causes them to precess like a gyroscope.
[34:49.200 --> 34:50.360]  So this should be spinning.
[34:52.040 --> 34:57.140]  And the frequency that they precess at is called the Larmor frequency.
[34:57.920 --> 35:04.700]  And it depends on the magnetic field strength and what's called the gyromagnetic ratio of the particle gamma.
[35:04.700 --> 35:07.400]  And all the particles have different ratios.
[35:07.980 --> 35:10.720]  So they all will resonate at different frequencies.
[35:11.540 --> 35:18.800]  So the Larmor frequency, we call that the resonance frequency or the center frequency of the scanner, because that's where all the RF equipment is tuned to.
[35:19.260 --> 35:24.400]  And the frequency is equal to the gyromagnetic ratio times the strength of the magnetic field.
[35:25.820 --> 35:29.660]  And for hydrogen protons, this is the gyromagnetic ratio.
[35:30.060 --> 35:33.080]  And you could see how the frequency changes at different scanner strengths.
[35:33.300 --> 35:35.480]  And the frequency goes up with the scanner strength.
[35:35.900 --> 35:39.380]  Generally, SNR tends to go up with the magnet strength as well.
[35:40.380 --> 35:42.570]  So this is the NMR spectroscopy.
[35:43.020 --> 35:44.060]  This is a frequency spectrum.
[35:44.060 --> 35:48.420]  And you could see how different molecules resonate at different frequencies.
[35:48.420 --> 35:52.140]  And you can look at the chemical composition of your sample.
[35:53.600 --> 35:58.240]  And one cool application of that is to detect counterfeit olive oil.
[36:04.480 --> 36:12.120]  So what we do is we hit the protons with a RF pulse, which also has a magnetic component.
[36:12.260 --> 36:13.580]  And we call that the B1 field.
[36:15.600 --> 36:21.980]  And it's actually, because it's an oscillating radio signal, it's rotating around the scanner at the same frequency as the protons are precessing.
[36:22.560 --> 36:24.880]  And it's always at 90 degrees rotating.
[36:25.100 --> 36:31.140]  And it knocks the protons 90 degrees away from the B0 axis.
[36:31.560 --> 36:33.640]  And now they start emitting radio signals.
[36:34.680 --> 36:40.200]  And then it also lines them all up in phase.
[36:40.380 --> 36:43.920]  And that's how we get the signal, because otherwise they would be all randomly aligned and we wouldn't get anything.
[36:44.460 --> 36:46.180]  And this is what the signal looks like.
[36:46.420 --> 36:51.680]  So as soon as you turn the RF pulse off, you get the signal starts to decay.
[36:51.680 --> 36:54.020]  So you can see it oscillating at the Larmor frequency.
[36:54.020 --> 36:57.700]  And then it's decaying back towards alignment with the B0 field.
[36:58.240 --> 37:00.700]  And this is what we call a T2 effect.
[37:00.880 --> 37:01.500]  I'm sorry.
[37:01.680 --> 37:03.140]  This is what we call a T1 effect.
[37:03.200 --> 37:08.200]  And then a T2 effect is the dephasing of the protons, also causes the signal to decay.
[37:09.340 --> 37:16.340]  And this follows what's called the block equations, which are T1 and T2 are the relaxation time constants.
[37:16.400 --> 37:18.140]  And those go into exponential functions.
[37:19.600 --> 37:23.680]  And so T1 is how quickly it realigns with the B0 field.
[37:23.740 --> 37:25.820]  And T2 is how fast the protons dephase.
[37:26.120 --> 37:27.960]  And T1 is a lot faster than T2.
[37:28.120 --> 37:38.120]  And we can adjust the scanner parameters to get different tissue contrast, because T1 and T2 are fundamental properties of the different molecules, just like mass or density.
[37:39.400 --> 37:41.640]  So we have all kinds of different RF coils.
[37:44.520 --> 37:46.900]  And today's imaging is a lot of parallel imaging.
[37:47.060 --> 37:51.600]  So we might have 12 coils or, you know, 64 coils in one, what we call a head coil.
[37:51.760 --> 37:57.380]  And we want to get the coils as close as possible, so that because the signal is very weak, that is coming from your body.
[37:57.640 --> 38:01.200]  So they might, instead of wrapping a coil around you, they might lay it on top of you like this.
[38:01.640 --> 38:05.180]  And the straps are just to keep the coil from shifting during the scan.
[38:05.180 --> 38:06.740]  So that's not nothing to be afraid of.
[38:07.580 --> 38:13.380]  And there's a lot of research coils, and they can be as simple as a wire loop, or they can be like fairly complex.
[38:13.680 --> 38:18.280]  This is for multinuclear imaging, or it can be tuned.
[38:18.560 --> 38:20.360]  The tuning can be adjusted to different molecules.
[38:20.860 --> 38:22.820]  So it's not just hydrogen protons.
[38:23.460 --> 38:24.700]  There's also a flexible coil.
[38:24.840 --> 38:32.460]  So if you have pain in your wrist, but it only hurts when you flex your wrist, maybe we can see why, but we can't do that with a rigid coil.
[38:32.460 --> 38:37.500]  So a lot of people are working on flexible coils, so you can get images in both positions and see the difference.
[38:38.920 --> 38:39.920]  This is kind of cool.
[38:40.080 --> 38:48.640]  It's an OpenCV hack to use a QR code to track RF probe, and you can look at the field from the coil.
[38:49.900 --> 38:54.460]  So the gradient coils for the spatial localization are really complicated.
[38:54.700 --> 38:56.560]  They have these really crazy fingerprint shapes.
[38:58.420 --> 39:00.540]  They are not superconducting.
[39:00.680 --> 39:09.320]  They're electromagnetic coils, but they are liquid cools, because they have very high current and voltage, and they're switching very fast.
[39:09.480 --> 39:12.760]  So both the coils and the amplifiers are liquid cooled.
[39:13.000 --> 39:18.300]  And this is, again, what causes vibration of the scanner, and that's where the MRI noise comes from.
[39:18.300 --> 39:20.300]  The magnetic gradients...
[39:20.840 --> 39:30.920]  So the B0 field in the magnet, if we're looking along one axis of the magnet, the B0 field is ideally very linear and very even.
[39:31.240 --> 39:38.300]  And because of the Larmor equation, the frequency that the protons are processing at should be the same in all regions.
[39:38.860 --> 39:43.760]  So that's not helpful for creating an image, because we can't tell if the protons come from here or from here.
[39:43.940 --> 39:48.300]  So what we do is we apply a slope to the magnetic field, and we change...
[39:48.840 --> 39:55.300]  By doing that, because of the frequency dependence, now the Larmor equation has a dependence on position.
[39:55.740 --> 40:05.640]  So now the particles over here are going to process at a lower frequency, and here they're still at the center frequency because we're crossing zero, and here they're at a higher frequency.
[40:05.640 --> 40:10.880]  And now we can look at the frequency spectrum of the RF signal, and now we have spatial localization.
[40:11.660 --> 40:13.220]  And that's the trick.
[40:13.340 --> 40:14.460]  That's called frequency encoding.
[40:16.020 --> 40:19.400]  And by doing this, we can select a slice with one gradient.
[40:19.620 --> 40:20.700]  I'm going to go through these quickly.
[40:21.800 --> 40:29.120]  And we apply different gradients in different directions to get the spatial localization, because we have a three-dimensional problem, so we have three gradients to solve it.
[40:29.440 --> 40:41.180]  And the phase encoding gradient has to be repeated multiple times, and that's where we manipulate the phase angle of the protons to get additional spatial information.
[40:41.420 --> 40:46.000]  But that has to be repeated many times to fill up the image, and that's why the MRI takes so long.
[40:47.180 --> 40:52.660]  And by adjusting the timing of the scanner, we can get different image contrasts.
[40:52.840 --> 40:57.640]  And you can see how here, these are T1-weighted and T2-weighted.
[40:57.700 --> 40:59.340]  These are all the same samples, so they all look the same.
[40:59.340 --> 41:00.740]  This is copper sulfate.
[41:00.840 --> 41:05.980]  These are LEGO pieces that they machined, custom-made LEGO pieces machined as fluid chambers.
[41:06.100 --> 41:06.940]  I thought that was really cool.
[41:08.400 --> 41:10.680]  And here, the oil is brighter.
[41:11.060 --> 41:15.420]  The copper sulfate and the oil are brighter in T1, but the oil...
[41:16.440 --> 41:23.060]  The copper sulfate doesn't show up at all in T2, but the oil does, and then the water is brighter than the oil in T2.
[41:23.060 --> 41:31.220]  So by adjusting the image weightings, you can get different imaging contrasts, which have different diagnostic purposes for the doctor.
[41:31.360 --> 41:32.500]  And you might do multiple scans.
[41:32.600 --> 41:33.900]  You might do a T1 and a T2.
[41:34.640 --> 41:37.900]  You might do other ones, because they each have different information.
[41:39.540 --> 41:41.480]  I'm going to skip through a little bit of this.
[41:42.240 --> 41:50.120]  So we acquired the image in K-space, which is frequency space, and it's a grid, like a matrix on a computer, or a 2D array.
[41:52.000 --> 41:58.380]  And we acquired the frequency information, and we acquire one line at a time using the gradients.
[41:58.540 --> 42:01.080]  And so this is why it's slow, and you have to fill up every line.
[42:01.120 --> 42:03.860]  And this might be like 256 by 256 matrix.
[42:04.000 --> 42:07.800]  So we have to repeat the pulse sequence 256 times to get one slice.
[42:08.720 --> 42:15.680]  And then we do the inverse Fourier transform to convert the frequency space, which doesn't look like anything, to the actual image.
[42:17.460 --> 42:26.820]  And if you're not familiar with the Fourier transform, if you're familiar with sounds, you can look at the low frequencies and the high frequencies, given the time signal of the sound.
[42:27.100 --> 42:30.580]  And, you know, you can boost the base, or you can boost the high frequencies.
[42:31.780 --> 42:37.300]  But this is the FFT right here, is dividing a time or a spatial signal into frequencies.
[42:45.070 --> 42:47.510]  So this is what the fully reconstructed image looks like.
[42:47.630 --> 42:49.750]  Most of the information is contained in the center of k-space.
[42:49.930 --> 42:55.590]  So if we ignore the outer part of k-space, I don't know if you can see, but the rest of it is grayed out, you get the low frequency.
[42:55.770 --> 43:01.630]  And now we're talking about spatial frequencies, so the black and white variations in the image.
[43:03.670 --> 43:05.670]  If we ignore all the...
[43:06.410 --> 43:11.210]  If we only look at the center of k-space, we get a low frequency image, which is kind of blurry, but you see all the structures.
[43:11.430 --> 43:15.810]  And if we ignore the center of k-space, we lose the structure, but we get all the high frequency details.
[43:17.450 --> 43:19.730]  So access is a big problem in MRI.
[43:20.310 --> 43:27.390]  We used to talk about portable scanners, and what they meant was a tractor trailer with an MRI in the back.
[43:27.670 --> 43:31.210]  And this is actually like what's used in a lot of more rural locations.
[43:31.930 --> 43:35.730]  They might have like a truck, you know, pulled up to the side of the hospital that's permanently wired in.
[43:36.250 --> 43:37.710]  And so what...
[43:37.710 --> 43:41.910]  Now there's more portable magnets nowadays.
[43:42.790 --> 43:43.230]  These...
[43:43.230 --> 43:44.770]  This one, I'm not sure if it's portable or not.
[43:45.390 --> 43:46.670]  But this one is portable.
[43:46.830 --> 43:48.070]  You can push it around the hospital.
[43:48.270 --> 43:55.370]  And it has this really funny barrier so you don't get too close to it because it has a permanent magnet.
[43:56.910 --> 43:59.970]  So the technology is really advancing rapidly.
[44:00.350 --> 44:02.750]  But again, like MRI is also very expensive.
[44:03.230 --> 44:05.170]  So this is where open source imaging comes in.
[44:05.910 --> 44:06.950]  So it started...
[44:06.950 --> 44:16.170]  I don't know exactly when it started, but one of the earliest projects was in 2014 and they were trying to build a tabletop educational scanner for less than $10,000 because you could buy it commercially for a lot of money.
[44:16.410 --> 44:23.890]  This has a one centimeter field of view with 50 ppm homogeneity and a 0.19 Tesla permanent magnet.
[44:24.530 --> 44:30.630]  And you might have several of these in a classroom so that you want to keep the cost down.
[44:31.630 --> 44:38.710]  Right now, the open source imaging project is working on a fully open source MRI that can be used for human imaging on a smaller scale.
[44:38.710 --> 44:41.650]  So maybe like a limb or a head or something like that.
[44:41.830 --> 44:44.970]  And they're actually seeking regulatory approval right now for human imaging.
[44:45.730 --> 44:53.770]  And this is what's called a Hallbach array where you use tiny little neodymium magnets and you put them in these little square slots.
[44:53.950 --> 44:55.010]  And this is a CNC machine.
[44:56.150 --> 44:58.110]  And it creates...
[44:58.110 --> 45:08.130]  By putting all the magnets in a very specific arrangement, which you need a computer simulation to design, you can get a very homogenous low field magnet.
[45:08.330 --> 45:10.930]  So this is like a mini MRI scan magnet.
[45:11.270 --> 45:14.790]  And they also added shims with sliding trays.
[45:14.790 --> 45:17.430]  So you can fix the magnetic field even better.
[45:18.970 --> 45:21.630]  This is one of my favorite projects on open source imaging.
[45:21.930 --> 45:31.950]  It's basically like the same as a 3D printer, but they put a magnetic probe on the end so they can measure the homogeneity of the magnet and create that B0 map that we saw earlier.
[45:32.890 --> 45:34.390]  And this is...
[45:36.210 --> 45:37.690]  Do not try this at home.
[45:37.850 --> 45:42.470]  And that's as much of a warning for me as it is for you because this is where I got that idea to try the stepper motor.
[45:42.470 --> 45:46.730]  Because I was like, oh, maybe I can put a stepper motor in the MRI room.
[45:47.010 --> 45:48.770]  So yeah, do not try this at home.
[45:49.170 --> 45:51.270]  But yeah, it was great for ideas for me.
[45:51.730 --> 45:54.090]  So you have to be extremely careful doing something like this.
[45:54.170 --> 45:58.830]  They put a really long rod on the end because they can't put this close to the magnet because it would just...
[45:58.830 --> 46:00.270]  It would fly in and get stuck and hurt someone.
[46:00.990 --> 46:05.270]  This is a 3T magnet and this is a 7T magnet.
[46:05.270 --> 46:07.070]  So this one you have to be even more careful in.
[46:08.970 --> 46:11.470]  But it's, you know, it's a really cool idea.
[46:11.470 --> 46:16.270]  And I never would have thought it, you know, especially being new to the field, I never would have thought it would have been okay to do this.
[46:16.390 --> 46:23.230]  But as long as you're super careful and you have the proper training, you know, you can, you know, it's pretty interesting what you can do.
[46:24.170 --> 46:27.510]  There's all kinds of RF amplifiers projects on there.
[46:27.590 --> 46:28.890]  So there's a lot of hardware projects.
[46:29.530 --> 46:32.030]  There's a lot of software projects and SDR projects.
[46:32.890 --> 46:35.710]  SDR is opening up like all kinds of new possibilities.
[46:36.570 --> 46:39.370]  And there's plenty of open source software on the project too.
[46:40.770 --> 46:44.430]  So if you're interested in getting involved, opensourceimaging.org.
[46:44.570 --> 46:46.290]  They're always looking for help and collaborators.
[46:47.530 --> 46:53.710]  And there's some great talks on MRI Together conference.
[46:54.210 --> 46:55.770]  All the talks are available on YouTube.
[46:56.090 --> 47:04.230]  And there's a really interesting talk from the National Radio Astronomy Observatory that shows the relation between MRI and radio astronomy.
[47:04.510 --> 47:05.990]  And that's also really interesting.
[47:07.330 --> 47:08.070]  So, yeah.
[47:08.330 --> 47:08.790]  Thank you, everyone.
[47:15.150 --> 47:15.690]  All right.
[47:15.790 --> 47:16.290]  Thank you.
[47:16.670 --> 47:19.290]  If anyone has any questions, he can't hear you directly.
[47:19.310 --> 47:21.170]  But I can try and repeat them.
[47:21.370 --> 47:22.530]  Or we can go up to the...
[47:22.530 --> 47:24.850]  If you want to go up to the front mic, you can speak directly in.
[47:25.130 --> 47:25.950]  Whichever you prefer.
[47:28.810 --> 47:30.010]  Yeah, I'll do my best.
[47:30.010 --> 47:38.600]  Without a high B0 homogeneity, what is lost in the image is the question.
[47:39.900 --> 47:47.780]  So going back to the Larmor equation, we get the spatial information from the frequency that the protons are processing at.
[47:48.520 --> 47:54.280]  And so we need the gradients to be very linear so that we have a known...
[47:54.800 --> 47:55.680]  Where did we go?
[47:56.540 --> 48:01.640]  We need the gradients to be very linear because the frequency has a direct dependence on the magnet strength.
[48:01.640 --> 48:08.400]  So if there's any variation in the magnetic field, then this line will get really funky and then we can't tell where the signals are coming from anymore.
[48:10.360 --> 48:10.920]  All right.
[48:11.040 --> 48:11.380]  Thank you.
[48:12.000 --> 48:12.820]  Another question?
[48:14.300 --> 48:14.760]  Yeah.
[48:19.800 --> 48:22.420]  Is there anything you can share with us about LibraHub?
[48:23.660 --> 48:24.680]  I'm not sure what that is.
[48:24.760 --> 48:25.020]  I'm sorry.
[48:27.820 --> 48:28.280]  Okay.
[48:29.600 --> 48:30.520]  One more question.
[48:31.620 --> 48:32.080]  Hello.
[48:32.360 --> 48:33.200]  I'm Xander.
[48:33.400 --> 48:34.080]  Thanks for the talk.
[48:34.800 --> 48:42.740]  I actually get a full body MRI every year through a company in the San Francisco Bay Area called QBio.
[48:43.020 --> 48:46.760]  And as you said, it takes about 50 minutes to do the scan.
[48:47.060 --> 48:54.060]  And I'm curious, is there any kind of physics-based impossibility proof around that time?
[48:54.060 --> 49:00.220]  Or could we conceivably do this in five minutes through some hardware innovations?
[49:01.340 --> 49:04.260]  There are a lot of hardware bounds on what's possible.
[49:04.560 --> 49:06.780]  It's basically how fast can you get your gradients.
[49:07.360 --> 49:09.200]  And there is a physical limitation on that.
[49:09.360 --> 49:16.560]  Because as the gradients get more powerful and as the switching gets faster, it actually induces what's called peripheral nerve stimulation.
[49:16.560 --> 49:18.980]  And it can cause, like, tingling in your arms.
[49:18.980 --> 49:20.640]  It's not dangerous, as far as I'm aware.
[49:21.200 --> 49:23.260]  But it's, you know, very unpleasant for the patient.
[49:23.460 --> 49:26.100]  So the speed of the gradients is a big problem.
[49:26.260 --> 49:31.020]  And the speed of the gradients depends directly on how fast we can acquire the images.
[49:33.500 --> 49:38.100]  You can actually, if you're getting an MRI, I've heard people that have been doing this a lot longer than I have.
[49:38.280 --> 49:45.380]  If you listen to the sounds, it can actually tell which type of pulse sequence is being run by the sound it makes.
[49:46.760 --> 49:47.200]  Excellent.
[49:47.360 --> 49:47.580]  Thank you.
[49:47.660 --> 49:50.820]  We have one question from our matrix chat from Odd Creation.
[49:51.100 --> 49:55.100]  You were comparing accelerometer-based measurements with cameras as the baseline.
[49:55.360 --> 49:57.020]  What's the advantage compared to cameras?
[49:58.760 --> 50:02.180]  The camera was kind of a secondary thing.
[50:02.500 --> 50:08.120]  Because I am very, like, I have a background in computer vision, so I'm really interested in using cameras for tracking.
[50:08.680 --> 50:16.260]  But we were just using the camera to verify the signal from the accelerometer because we can't program the patient's motion.
[50:16.460 --> 50:18.980]  The patient can move in any direction at any time.
[50:19.080 --> 50:21.860]  So the camera was the only thing I could think of.
[50:22.020 --> 50:30.200]  Because if we had a sensor that could do this already in the MRI, then I wouldn't need to do my job in the first place.
[50:30.820 --> 50:33.980]  So the fact that those sensors don't exist now is what we're working on.
[50:34.160 --> 50:40.420]  So the camera was just to verify it because we have no ground truth for patient motion.
[50:41.920 --> 50:42.400]  Excellent.
[50:42.520 --> 50:45.040]  We have one more question from the matrix chat.
[50:46.380 --> 50:48.260]  What are the image formats like?
[50:48.400 --> 50:51.440]  Can I make a 3D model of my own data without proprietary software?
[50:52.560 --> 50:53.120]  Yes.
[50:53.120 --> 50:54.680]  It's called DICOM format.
[50:54.860 --> 50:57.100]  And there's something called PI DICOM in Python.
[50:57.840 --> 51:00.880]  And, yeah, you should be able to import DICOM images.
[51:01.320 --> 51:06.480]  And I'm not as familiar with that side of it, but that should definitely be possible.
[51:06.580 --> 51:10.060]  There's also plenty of free DICOM viewers online.
[51:10.060 --> 51:14.160]  There's a program called 3DSlicer that can also import DICOMs.
[51:16.420 --> 51:16.900]  Excellent.
[51:17.000 --> 51:18.320]  We got one more question for you.
[51:18.660 --> 51:19.140]  Regarding,
[51:22.670 --> 51:26.410]  could that be taken, could that be corrected for by overlapping your slices?
[51:26.730 --> 51:35.450]  Or are the slices infinitesimally small that you can't reference the next previous one, next or previous one to it, if the patient was to move between the slices?
[51:36.050 --> 51:36.530]  Wow.
[51:37.590 --> 51:38.690]  I heard it.
[51:38.930 --> 51:39.030]  Okay.
[51:42.430 --> 51:44.330]  So each slice is acquired one at a time.
[51:44.430 --> 51:45.230]  There is...
[51:45.230 --> 51:52.190]  I went through the most basic format for this, but there is what's called 3D imaging now.
[51:52.350 --> 51:58.050]  And I'm not really familiar with how it works, but you can acquire a 3D image in one scan, as far as I understand.
[51:58.430 --> 52:03.550]  But because of the limitations on the gradient speeds and how fast we can acquire the slices, if there's any...
[52:03.550 --> 52:06.510]  Like even a heartbeat motion can throw it off.
[52:06.630 --> 52:12.310]  And it's also like the motion of the blood through your veins has a magnetic field.
[52:12.450 --> 52:13.710]  So even that has an effect.
[52:14.850 --> 52:15.390]  I'm sorry.
[52:15.490 --> 52:16.710]  I'm not answering your question, though.
[52:19.930 --> 52:21.410]  Thank you so much for the talk, Doug.
[52:21.510 --> 52:22.470]  We really appreciate it.
[52:22.790 --> 52:24.490]  Thank you for the audience for being here.
[52:28.920 --> 52:29.320]  Yeah.
[52:29.440 --> 52:29.780]  Thanks, everyone.
[52:30.460 --> 52:36.160]  And if Doug has any information or documents he wants to share the presentation, you should be able to find it.
[52:36.160 --> 52:39.760]  He can post to the Matrix chat for this talk and you will be able to see it there.
[52:40.620 --> 52:41.960]  Come back at 3 o'clock.
[52:42.060 --> 52:49.100]  Our next talk will be Secure Cell Phone Communication, Mission Accomplished, or Popular Delusion by Dr. Nick Germain.
[52:49.300 --> 52:50.860]  So we hope to see you here in 10 minutes.
[52:51.520 --> 52:51.920]  Thank you.