 Good morning or only just good morning. Welcome to Stage B. The talk now is by Dave Harvey. It's on digital medical imaging. Just a couple of things before we start. Firstly, if you've got some free time today or even tomorrow to help with Teher down, we would really, really appreciate some volunteers. So if you could spare a couple of hours of your time to volunteer at something, anything on site pretty much, it'll be greatly appreciated and you might get a free meal out of it. Also we've had a wallet handed in. yn gweithio'r gweithio'r ddyn nhw, Aberthrack Derham, yn gallu gweithio'r gwlad yn y gweithio, mae'n amddangos ar y rhanodol. A yna'r gweithio'r gofal o'r ddyn nhw'n gallu gweithio'r gweithio. Fel hynny, rydyn ni'n ardal i ddod. Ddod. Yn y golygu, rydyn ni'n gweithio'r ddyn nhw? Rydyn ni'n gweithio... Ond mae'n gweithio'r llaw o'r llaw o'r 1-hour. sy'n dwy'n dweud yn gyfwysig eich llwysoll, yn gyflwytaeth gyrnodd i ddweud yn ythafol o'r tynnu'r lleol yn y lleol. Felly, yna'n mynd i'w ddweud i'n ddych chi'n golygu, yna fydd sy'n mynd i'r ddweud y dystod o'r cyfnodd, y dywedodd arllun y cyfrifodol, i'w dweud o'i cael ei gweld o'i ddechrau. Ond yw y gallwch eich gweithio'r ddechrau, yw i chi'n gwneud, yw ymlaen i'w gwellio. Yn ei fod yn ymwneud i chi'n gweithio. Yn ymlaen i chi'n gweithio, ac yw'r gweithio'n gweithio. Yn ymwneud i chi'n gweithio, ymlaen i chi i ymwneud i'r pattern sy'n fyddfa i weld ein gweithio, ymwneud i chi, mae'n dweud i gweld i'r scan Yn ymweld yn ymweld os ysgolwyddiad yn gyffredinol. Rydym yn ymweld o'r ffordd ar y diolch. Yn ymwybodol yn ymdweud, mae'n gweithio'r ysgolwyddiad yn gyffredinol o flwyddyn, yn ymdweud hynny o'r gweithio mewn cyfnodau o'r cyfrwysau. Yn ymdweud, mae'n gweithio'r ysgolwyddiad ym MRI. Mae hyn yn gweithio yn ddechrau o'r dweithio, a mae'n ddim yn gyfweld i'r ddechrau. Mae hynny'n gweithio yn gweithio. Menonwyd yn ni o gyfemin a gweld gen i mi ddweud y gweithio i'r dweud, neu halu ddweud â'i hwyl ar gyfer, a chwn ni o ddaw ni'n ardal ddim yn cael cyfemin a gweld digwyddgau i gweithio. Mae gymryd ditudonol eich ddwylai ar gyfer ond yn ag仗. Ond yna'r cyffin arall darllen o ddweud hynny oherwydd rhai o phobl i ddim yn nadir i ddim. Rhai o fyddechrau, rhai o'r ffordd, rhai o'r ddweud o'i ffordd, mae'n gweithio'r bobl i ddim yn rhoi'r llyni i'ch gweithio'r gwaith i gweithio'r cymhwyl gael y cyfnodol. A'i gweithio'r gweithio'r cyfnodol? O'r siwr, mae'n gwaith i ddim yn gweithio'r cyfnodol. Felly mae'r drwy'r cyffredinol wedi ynnu 80-87. Felly dyna'r cyfnodol, mae'n rhaid i'ch ei wneud i'w meddwl i'ch dod i'w meddwl i'r gyfer i gyfer MFC i'r meddliadau. Mae'r gwaith ymlaen i gynnyddio'r gwaith ar y gyfer 9 ymlaen, yn ymlaen i gyrsdyn ni'n 133 oes ar y gwaith, yn ym 80 oes yn y cyfnod, yn ymlaen i gydag i'r ddweud o'r ddweud o'r ddweud i'r ddweud yma ymlaen i'r ddweud o'r ddweud. Mae'n ddweud i'r ddweud i'r ddweud, ac mae'r ddweud o'r ddweud. Felly dr Harvie, mae'n gwaith i'r ddweud o'r ddweud o'r ddweud. Er mynd i gyd achos ymateb yn ei bod yn ymdillaf gyda'r gweithu y gysyllt y Llywodraeth Cymru yn ei ddefnyddio'r hosbydl? Mae gynnal yw'r hunain. Alg ymddangodd o'r cymdeithas ar y cysyllt, o'r ffordd, yw rwy'n credu. Mae gennym eich bod a'i ddweud y cyfrannol a'r dda, mae'r ddechrau yn adshodd hwn fel ddoedd diwrnog yw'r meddliad. Epoch oedd gwneud hynny, ei gwneud yma ar gyfer 6 oed, yr ad Imaginaid, ynghylch yn Ysbyn ar gyfer o'r gweinyrch, ac yn ochr i'r cael eu dweithio'r cyffredig yn hyd yn gweld y dylai'r cerdd o'r newydd o'r newid yn cyfwyr llythodi'r provinces a beth dyna'r cyfrifun hwn i'r llwg iawn. Mae ddechrau pob i bobroio, drwsgaith, a dweudm yn mynd i'r ran o'r gwarn i'r cyfrifun. Mae'r directio'r company o'r llwg iawn. Mae hi'n iddynt yn unrhyw ddifigol yn y dyfoi, Felly mae'n ddweud i'n ddweud i chi, fel y byddai'n ddweud. Mae'n ddweud i'r ddaf yn ymweld i'r gweithio cyflwyno. Felly, ar y llai 14 yma, mae'n ddweud i'r drosbwynt yn cyfrannu gwahanol. Mae'n ddweud i'r ddweud i'r ddodol i'r ddweud i'r meddwl wedi'u meddwl, ond rydw i'r ddweud i'r ddweud i'r ddweud. Yn ynik yn llawer o'r gweithio cyflwyno ar y defin 빼ico. Felly y mynd i hardni caj placesio, mae nhw'n ddu i ni weld, ac mae socialau drws arnynd i gweithio a demand questions ar copi gweithio ac fel ddweud i'r gromabu'u cyngorledig i'r ffordd phoedd side b uncertain. Mae hyn yn dal wlad wlad yn ôl yn yrod i'r meddwl, aeth ni'n shaifi inequalitya petr ac yn yng白frannu. dweud yn awr. Fionar fel arddangos yn llwyddo arfer. Felly, sy'n fyrddwch chi dweud y dyna cynsgops appearsbwynt o'r corffmanau. Mae hyn yn cynnig y cyfre, sy'n gweld y system yw'r ardal fynd i cyfwyntiau. Mae'n byw'r ysgrifennid ysgrifennid yw'r RWI yn y Llywodraeth Merthyn Fylltwladol Cymlachau yn dda'r hynny, ond mae'r ysgrifennid, mae gennyn nhw. Mae gennych, roeddwn ni'n gallu gwirio ysgrifennid yn y gweithgaredd yma, yn gallu ei wneud i chi i gael 60% eich bod yn ei gawrbyn ddiolch, a byddai'n gweithio-i gweithio'r dweud i'r buswch ymlaen a byddai'n gweld a ddweud i'r buswch gael. Rydyn ni'n gweithio'r ddweud o'r byd oedd oedd yn gweithio'r beth ac yn gweithio'r beth ac yn gweithio'r beth, ac yn gweithio'r beth Ac rydyn ni, os ydych yn gweithio'r buswch i ddweud i'r ddweud i'r lles fatho'r sydd ydych chi'n clyw o hyfforddiol. Felly, mae'r mynd i gweithio i ddweud yma rhywbethau dyma, sy'n gyflym o'r cychwyn yn yw'r cychwyn. Roeddwn i'r cywbeth gweithio, mae'r ddweud y 5 mae'n ei ddim yn ei dweud o'r cychwyn a phobl o'r cychwyn yn unedig am ddod o'r hyn o'r cychwyn o'r cychwyn i'r ddweud. Dyna, ysgrifell hwn yn cyntafaf yn gweithio'r ffaserdog bwneun ffilm iau gweithio'r eu bwneun, hynny maen nhw'n iawn pleunio'r bwneun, rhai eich gweithio eich gweithio eich bwneun sy'n iawn gwneud o bwneun gwahanol a'r ffotonau sy'n dechrau i bwneun. Yn ymellai'r ysgrifell, mae gwneud yn wneud i gwneud eich radiograff, mae'n bwneun eu gweithio eich gweithio'r bwneun, o hynny maen nhw'n iawn llwyngau. a dweud yn ymweld 10, 15 oed i wneud amser y gweithio. Mae'r blwyddyn yn gallu glywed o 19th gennyddol. Mae'r blwyddyn yn gallu gael'i gydag ffilm o'r gweithio'r gweithio ac mae'r gweithio'r gweithio'r gweithio yn 35 x 43 cm gan unig o'r bethau achos. Mae'r bod yn eithaf dros gweld y dweud. Mae'r rhai gweithio'r gweithio. Mae'r riddyl yn gweithio'r image yn gweithio. ac mae'n ddaeth yn cymdraith o'r prosesor, yn loes cinnwys mewn ddeithas o ddeithas yn ddeithas yn ddeithas o ddeithas. Mae hyn o ddeithas yn gweithlabol, nad yna gweithio eich 25 ysgwrs, wedi cael ei fod wedi'u ddweithio, a'r wneud o'r prosesau i ddaeth, i ddaeth a'r ddeithas, ac mae'r ddeithas yn ddau'r ddeithas, felly mae'n ddaeth yn olygu'r ddeithas. Ond o'r ddeithas, ddweithio'r ddeithas sy'n gorffodio negol ychydag. Fe synodd i gynnig sy'n bob yn ei chymreu i'r ddweud oherwydd hwn yn ei cwm, i'n ei fod yn rechydig mae'n meddwl, ychydig ychydig oherwydd mae eich cwm yn ei chymreu, mae hwn yn meddwl i'ch ddweud, chymreu hwn yn twfyn i fynd i'ch ddygu. Y cerdd gen i'w deud yn mynd iddo i'r ddwylo, mae eich Llyfr cynghydwyd CBCD sy'n gweld i'ch ddweud o'r ddweud, wirer ddisgoel canfysag hyd wedi Cymru. That's very, very inefficient to try to do that because there's not many things which can actually record photons and have yet heavy nuclei in them to stop them efficiently so what you actually have which is what I've drawn as this black line on the diagram is a toss power essence plate and as each X-ray hits it it produces thousands of light photons and if you look at the relative energy of X-ray photons vs LIGHT energy photons ar gyfer citing, yw'r gwahod, ac ydw'r gweithio y gallur a'r chwarae unig yn lawer o'r tyli ac yn gweithio'r gweithio'r gweithio'r cerd對ion. Mae'n cyfeirio'r cerd wedi'u ddau'r gweithio'r gweithio cyfnodr yn iawn gyda'r gyffredinol newydd o mynd, i ddau'r cyffredinol gwahanol yn ddod, a'r craswn cymaen nhw. Rydych chi'n meddwl ym hyn, mae'n gobeithio'r gweithio'r gweithio'r gweithio'r gweithio. a'r syniad rhywbeth, ond y gallwn y rhan o'r drwntaf o'r amser. Mae'r blwysig yn ddweud â'r dweud wedi bod yn oed yn gwneud ar y ddweud. Mae'n ddweud a'r rhan o'r cymdeithasol gennynnu'n ddweud. Ond, y tro o'r ffordd, mae'r ffordd yw'r ffordd o'r ddystynwysgau. Ie, mae'n holl o'r ffordd o'r ffordd. Mae'n ddweud ar y ffordd, ond, ond, oherwydd, wedyn yn ddweud y ffordd. Felly mae'n ddweud y stwch ddimensial a'r ddweud yn dweud y ddweud. Mae'n ddweud ychydig. Mae'n ddweud yn ddweud yn ddweud yn ddweud, ddweud yn ddweud yn ddweud yn ddweud. Mae'n ddweud yn ddweud yn ddweud. Felly, yn y ddweud, y 60s a yn 70s, unrhyw bryliant yn y ffyrdd yn cyfnod, sy'n gyfnodd yn ein bod yn gofyn ni'n ddweud. Mae'n ddweud yn ddweud yn ddweud yn ddweud yn gyfnodd y project ym EMI. Felly, mae'n ddweud yn y same EMI that produce the Beatles records and worked out a way of taking hundreds of images from different directions around the patient, recording those digitally, feeding them all into a computer and when he first did it in the mid 70s, you then waited about three hours for the computer to crunch the numbers, Felly, mae'r cwysig ychydig yn fwyaf y gallu cyfnodol sydd wedi'i gweithio'r cyfnodol. Fyddaf yw'r ychydig yn y cyfnodol, yn y cyfnodol, o 64x64 o ddweud o ffawr. Fyddaf yw'r cyfnodol, o ffawr, o ffawr, o ffawr, o ffawr, o ffawr o'r cyfnodol a ffawr o'r cyfnodol ar y cwysig ychydig. Mae'n gweithio'r cyfnodol yn fwyaf yn ddweud. Hydyw'ch have to solve them! And they were solved initially, iterativily, which understood why it took hours and hours on the computers available then. Subsequently and in the meantime most of the work has actually been done using a few methods which yes, I won't go into here. There's enough for you to battle you all later when we come to the MRI. But interestingly in the last five years or the same people have gone back to doing a few methods. .. ac mae'r amgueddur gennu ded Flu nesaf i fodwch a gweithio gael i'w cael y maen nhw yn dyn nhw. Mae'n gweithio'n gweithio'n gweithio'n gweithio oherwydd mae'r amgueddur rei. Oherwydd mae'r våce o'r alternativ i Gwithi Gwithoedd maen nhw'n gweithio'g wnaeth. Mae na fyddwch yn gweithio'n gweithio'n gweithio'n gweithio rhaid i gyd. Mae'n gweithio'n gweithio'n gweithio'n gweithio gwyll retainingu. Ac mae'n go iawn ac mae'n gweithio'n gweithio. I spent hours on PowerPoint making this work, so I hope it does go as intended. Yay! There you are. Thank you. Thank you. So that's how CT works. And of course, when it's done it all, by the way, the green blob in the middle is the patient, in case you haven't worked out. You end up with a picture like that, which is actually a picture of my father's brain. And you'll be seeing more of that as we go through. I do keep this very much family-related, because I know I've got permission to use those images. Right, ultrasound. One of the non-invasive ones. This one's not going to be registering on the total population radiation count as shown on the dashboard at this event. I'm not going to go into a great deal of detail on ultrasound other than to say that it's basically high-frequency sonar. You're running signals in the few megahertz range. And I say that deliberately, because, in fact, the modern machines actually use swept frequency ranges, because the higher frequencies give you better resolution close, but don't go very deep. The lower frequencies don't give you as good resolution, but they go deeper. So it's quite complex modern ultrasound machines. But basically, it's just like sonar. You send out a pulse, you wait and see where it comes back and where it comes back from, and you make a picture. I'm sure you've all seen ultrasound pictures, antinatal ultrasound scans. Interestingly, from the point of view of what I'm talking about today, which is how images get passed around, the captures, the pictures you actually take of ultrasound are almost useless. Because you don't have a standardised system for where the probe is and recording it and so on. The only person the picture really makes sense to is the person holding the probe, who knows what they're looking at. We have to record them to keep the lawyers happy. Medical legally wouldn't be allowed to say, oh, I've done it, here's my report, and not have something to go with it. But actually the images are of very little use. So quite a few people now are actually recording it in video and if you want to, you can actually encode it up in MPEG and I'll talk more about how that works with DICOM later. Like almost every part of radiological imaging, the images you get back are actually single channel. They're monochrome, grayscale. But that doesn't stop people trying to combine different sources of information into the same picture if they want to. So what you can do with ultrasound, because as well as actually working out what signal comes back from where, you can actually measure its frequency. And with a bit of work, you can then use that to actually work out the speed of what you're looking at via the Doppler shift. And that then can get overlaid on the image as a colour signal, which is incredibly useful when you're looking at the heart because it can be very difficult when you're looking at a valve which should be that wide and has been narrowed by 10% or 20%. Trying to actually get accurate measurements and that is almost impossible. But because the flow through it has to be fairly constant or the person wouldn't be alive, if you've dropped the diameter by 10%, you've increased the flow speed through it by 20% because of the square laws. And therefore you can really measure that accurately. So this is a sort of ultrasound image you get. This is an antinatal ultrasound. This is actually my own twins. And if you look hard at your left, you will see the head of my daughter. And on the right, you will see my son, who's 18 a week today. And you can tell it's my son because he looked very, very carefully. Have you got the mouse up there? That's one leg, that's the other leg, and there's his balls. And he's sitting here looking embarrassed. Right. So that shows how old this technology is. He's 18 next week and this is over 18 years ago. But that's actually, funnily enough, the capturing of these ultrasounds, which is of very little use, but was using vast quantities of film, costing us £40,000 a year, but incredibly eco-unfriendly in terms of all the silver, et cetera, was actually the reason why I started playing around with digital imaging. In these days, this wasn't even direct digital. This was literally a video capture of the only output which the ultrasound machine had, which was a video. I put it into a video card, added the demographics, and stored that as a magical image. It saved our hospital £40,000 a year. It then caused them eventually to lose me and I decided that was far more fun than actually taking the pictures. On the other hand, here's the other source you can get where you've got ultrasound, where you've got Doppler, sorry. Again, this is only really two-channel information. There is the reflectivity information, the grayscale, and there is another independent variable, which is the flow. But traditionally, if you want to show two things, you don't just use two colours. You use one as grayscale on top. So it's a nice picture showing the heart with measurable flow. Right. Positron emission tomography. How many of you heard of positron emission tomography? Okay, most of you. Sorry? Good. Well, it's actually, it is a wonderfully cool imaging method. It was invented about 20 odd years ago. There was only one slight problem. For the first 10 years or so they had it. They knew it was cool. It was wonderful. They knew it was technologically marvellous. Nobody could think of a use for it. They just hadn't worked out what you could do in terms of what substances you could attach the radioactive tracer to. I'll talk about how that's used in a minute. What you could attach it to in a means that would provide useful information clinically. They have got there now. They've got various markers which they can use, which therefore guide your radioactive tracer to tissues which are particularly some cancerous tissues. You've got a marvellous way of localising where the tumours are because you've got a material which is guided there. What's brilliant about positron emission tomography is that everything else we're doing here is working on effectively emergent properties, stochastic effects. You're averaging out all these things that are different happening. What proportion of your photons get through? You've got quantum noise, how many are you going to get through, how many aren'ts and so on. Positron emission tomography, you are actually recording individual radioactive events. What happens is that if it's a positron emission, beta plus decay, what's emitted obviously is a positron. What happens when a positron meets an electron, when it doesn't make more electrons, they turn into a pair of photons. Now, admittedly, the majority of those photons are going to be out of the plane of your scanner and you'll never see them and you'll never pick them up. But for those that are within the plane of your scanner, as we're showing here, if they both hit the detectors within a very short period of time of each other, you know that the emission must have been along that line. You also know the relative timing with which it hit those two detectors and you know the speed of light. Therefore, you can actually place exactly plus or minus it's a few millimetres of centimetres. It's not one of the highest resolution imaging formats we've got, but you actually know from that single event where in the body it was. Obviously you then collect a few tens of thousands of these. There are practical challenges with PET largely due to the fact that the material used has got a half-life of only a few hours and you can only make it in a cyclotron. Therefore, and by the time you've taken into account all the extra time after injecting is in the patient for it to get around the patient, etc. The challenges of actually getting your isotopes made, isolated, produced to pharmaceutical standards, shipped around the country, injected into a patient and then the patient imaged when your half-life is, I can't remember the half-life of 15-nose oxygen. It's a few hours, it's very short and that does limit the availability of PET to major centres. But it's relatively low resolution so what you actually tend to do is a bit like I was saying with the ultrasound with the Doppler is that you actually overlay it on top of other images and typically PET and CT are used together so the CT provides the high resolution picture to show you exactly which bit of the body you are and the PET then provides the functional overlay to show which bit of the which bit of that tissue is taking up the isotope which according to what the isotope is you can then tell about it. Now MRI, sorry. Sorry, where have I got? Position. Oh position, yes sorry. I should be positron. Correct. Thank you very much. Absolutely yes. I love nitpicking at the people so I can't complain where people do it to me. Right. We're now on to MRI which I'm going to go into in some detail because let's face it, it is the only modality that actually uses a strong electromagnetic field. So I have to talk about that here really. And we are talking very, very strong magnetic fields. Magnetic fields are measured in Teslas. Now I realise most people when they talk about DMF, when they talk about a Tesla will be thinking something like this. But no, that's not what we're talking about here. The Tesla is the unit of magnetic field strength. And to give some ideas, there's a magnetic field fraction of a micro Tesla. Fridge magnets are a few milli Teslas. High power motors can sometimes have a few Tesla magnets in them but they're tiny. An MRI machine needs to be big enough to take a whole patient in it. It needs to have a field. They're between about one and five Tesla. And they need to be incredibly homogeneous to within a few parts per million. So that's how they're actually built. I was going to try to show a real picture. I think the cutaway gives a much better idea of just the scale of these things. And you realise why they cost a few million pounds each. Because not only do you need to build this, have it this accurate, you also need to keep the whole thing for the whole of its life pretty much at superconducting temperature so it's full of liquid helium. Because you could not make a resistive magnet that was this strong and this even. So they're all superconducting magnets. So what is nuclear magnetic resonance? Which is what? Magnetic resonance imaging depends on and I'll talk about the distinction in a minute. Whenever you've got a magnetic field any nuclei with an odd spin spin around an axis parallel to that field. Now, for those of you who have got degrees in Theoretical Physics please accept my apologies I'm doing a gross oversimplification here. I'm not distinguishing between the spins of individual nuclei and the overall resulting field. This is massively simplified. I hope it's good enough for this talk. If anybody wants to shoot me down on my physics later please do so. But you end up with these nuclei and we're talking hydrogen for imaging. They're either aligned with the magnetic field we imagine there's a magnetic field going up and down on this slide or they can be persuaded to flip into the other field by a radio free by photon of exactly the right energy. Which also happens to be various works of quantum mechanics photon which has the same frequency at which it is processing because it's like a spinning top. It's spinning itself but then what we're talking about isn't it spin speed, it's procession speed it's the speed with which it goes round and round and that speed and this is critical to the whole of the rest of the imaging is dependent on the magnetic field strength. It's about 47 megahertz per tesla. So which actually puts it roughly on a 2-3 tesla magnet. I mean you're talking 100-250 megahertz you're actually talking roughly in a sort of standard radio band. I've seen badly screened MRIs have all sorts of artefacts which have turned out to be due to local radio sources or the local taxi company because that is the sort of frequency band we're talking. Hason do add of course non-ionising therefore you're not got the radio you've you've not got the ionising radiation effects. So when the frequent when that energy is applied it does this it flips but that's its higher energy states like most things in the world when they're in a higher energy state what they like to do is to flip back. So if you put a photon in to make it flip what happens when it flips back you get a photon back out and that is basically nuclear magnetic resonance. Chemists do it for years any of you have done a level chemistry in the last 10 years or so would have come across it they do it in terms of parts per million shifts and that's called magnetic resonance spectroscopy that is not what we're doing. This simple NMR is great for a test tube full of stuff you put it in there you blast it you get the signal out and that's fine but it's not much use knowing what the signal from a whole person is. We need to actually know what the signal is and how much material there is and again I'm not going to go into details but whether it's fat or water or whatever actually has an effect as well on the different types of signals. So you're not just seeing how much there is but let's assume for the moment that we're just seeing how much there is of a particular type we actually need to be able to resolve it. That's the difference between nuclear magnetic resonance which is the overall effect and magnetic resonance imaging which is where you're actually resolving it into a picture and in order to do that of course you need to have something which is different about the different directions so it was on the diagram but I didn't talk about it. When you have your MRI scanner as well as the main field which is this incredibly powerful field, there are also three sets of what's called gradient coils which can change the field by about 1% it's typically a few tens of milli Teslas from one end of the patient or one side of the image or front and back it's three so you can resolve into three dimensions and I'm going to call them XYZ. In fact you can by using combinations of them and this is one of the great things about MRI compared to CT because you can use them in any combination just with a simple vector of multiplication you can actually image in any plane so whereas a CT you have to image axially the other old name for it is the CAT scan computerized axial tomography because that's the only way you can image see an MRI you can image me this way this way this way or obliqually if you want to and that's very easy but for the time being we're just going to assume we're talking XYZ as fixed dimensions and of course these field coils produce a field which is in the same direction as the main field but they're changing all the time but they're producing a field which is interacting with the big long strong field can you imagine how much force there is on these coils you put one magnet inside another and then change it rapidly backwards and forwards that's going to have some rattling on it anybody who's ever been in an MRI scan has probably been given ear plugs that's a very good reason because the actual vibration on these field coils is very very strong and very loud but what I'm going to talk about is how you actually do this work and I one of my colleagues down here looking at the side was very proud to point out to me said am I going to mention Nottingham and he was very surprised to feel that I am because this work was done in the early 70s in Nottingham another UK Nobel prize on top of the one that Godfrey Hounsfield got for CT because of the work I'm now going to explain into how you actually resolve this signal into three dimensions lots more animated power points so resolving in one dimension I'm calling them XYZ but it really doesn't matter you can just call them A dimension another dimension and A third dimension but let's stick with XYZ for now it's actually quite easy because of the resonance effect of the gradient and then you put in a pulse of radio frequency energy then natural frequencies if it's what I'm showing here with the blue this is what they would be this is what they would be processing at are different because of the gradient obviously exaggerated effects on here so if you have your external radio frequency energy and you put it in there's only the one slice of the patient where you're going to have the resonance so only the protons, only the hydrogen nuclei in that slice are going to be excited the rest are going to be ignoring it, it's like you've got neutrinos flying through so they will absorb the energy they will become excited so everybody with me so far we've now got one slice let's call it a slice around my middle and call that the X before I go on to the X bit I need to do a little bit of a tutorial because it's important for the rest of it I'm going to talk about Fourier transforms anybody here not to know what a Fourier transform is? enough of you to make it worthwhile if you look at your music system and if you've got a fancy music system especially an old one from the 70s or 80s it may well have a spectrum analyser display on it something that actually shows you all the different frequencies it's actually a very crude simple one most of them they do it by octave what have you got in the super bass down in the 20 hertz range what have you got in the middle middle range around a couple of hundred hertz what have you got up in the high frequency high hat but the principle of what you do with a Fourier transform is just the same any signal can be broken down into a series of sine waves and those series those series of sine waves when put back together with the correct phase actually then can reconstruct exactly the original signal subject to some bandwidth of things but we know that from the main the basic point is that you can look at the frequency way of looking at things or the time waveform looking at things as two different ways of looking at the same data so just like if you were to take a music sound and you can you could either say here it is it's with this shape repeating or you could say ah yes that's a signal of 14 hertz a signal of 32 hertz and a signal of 53 hertz in the following ratios and there are different ways of looking at the same and what we're going to do with MRI it depends fundamentally on this idea of a Fourier transform so we're now going to take us red that we've resolved the X dimension simply by what we have excited so in terms of what your screen the X dimension is this way what's behind the screen here didn't get excited in the first place and doesn't play what's along the front row here doesn't get excited and doesn't play but what's in the plane of the screen is the ones that we've actually excited by putting the beam in having applied the gradient during the time we put the beam in we put the energy in so we've now got these 16 identical pixels let's call them okay there's thousands in the real image but let's assume we've got these somehow we have to make them all different such that they're going to be solvable because it basically boils down to simultaneous equations again so it's actually quite easy to do one of the dimensions which we're calling Y because what you do is that you actually change the Y gradient during the time when you're getting the echo you apply one gradient while you put the energy in you've spun up just that sequence then while you're getting the echo back out you apply a gradient the other way which changes the speed they're spinning so what you get out is a series of frequencies and each frequency tells you where in this case horizontally what we're calling the Y direction where that signal is coming from it really didn't take a great deal of maths and work for people to work out how to do this here's the difficult question though you've used one of your gradients during excitation you've used a second one of your gradients during um readout how do you actually try to get resolution in that third dimension how can you actually get a difference between the differences along the Z axis and this is what the guys actually got their Nobel prize for and I think they deserved it so there we are sorry I'm slightly ahead of my talks so you've used the X gradient during excitation the Y gradient during readout how can we use the Z gradient the answer is that you use it in that gap in between but what does that change how can you use that information well what it changes is the phase if you've speed something up for a little while if you've got two people walking along at the same speed one of them speeds up a little bit and then they go back to walking at the same speed they're not next to each other again because the one that's sped up for a while has moved ahead a bit and in things that are spinning that's called a phase change it's gone however many degrees in advance so what you can do with the Z and what you then do I'll show you this for a moment is you've got them all spinning you then and you have to watch this carefully because it's subtle you then apply the Z gradient change which does this did you all notice what happened they sped up by different amount just for a short while such that if you look at it there's a 45 degree phase angle between each row when the top row is vertical the row beneath it is pointing at half past 1 the row beneath it is pointing at 3 o'clock and the row beneath that is pointing at half past 4 so we've shifted the phase by putting in this intermediate then of course at readout time you apply this business about changing the speeds and what we've now got is all 16 are doing something different and that's the fundamental thing you need in order to be able to solve the maths yes this was fun doing it in Powerpoint I spent far too long on this talk so what you can't do is you don't have enough information just from one set of readouts because you've only got that one signal with that one set of Fourier so what you actually do is you have and this is why MRIs take a long time anybody who's had one will know that you're having to sit there for several minutes we're going all sorts of wonderful noises that they make and I won't try to imitate them all you do it multiple times with a different amount of the phasing coding, the Z every time and you build up a complete and you have a series of these lines oh how do I I just went backwards so each one becomes a line you then for as many rows or columns you need for your image you need to do that number of what's called phasing coding steps and you end up producing a picture like this and this is a real picture and then I'm afraid I have to quote Mr Ben here you do a two dimensional Fourier transform and as if by magic a picture appears and that is literally what happens at this stage you've taken the data you've done a Fourier transform which gives you the Fourier transform of the image you do the different phasing encoding steps and out comes that picture and I think the actual maths behind that and the simplicity of it when you actually look at it in the end compared to what had to be done with the CT it has a marvellous symmetry simplicity and as we pointed out and why they got the Nobel Prize utility as well so that's how you get an MRI picture so where am I so you've made all these different images you've possibly got a semen CT scanner a GE MRI scanner an acuson ultrasound machine and a codec DR direct digital radiography how are you going to look at all these images without having to run around and look at all these machines and this is what I'm going to talk about for the next part of my talk which is the standard that's used for actually exchanging these images once you've made them and effectively it's a bit similar to what you have with JPEG you know everybody's digital camera whoever's digital camera it is will make JPEGs okay there were people who tried to do other things a few years ago but they pretty much died out so and it's actually JPEGs and DICOM is really more closer to TIF but we'll ignore that for the moment but the point is that it's not just the image it's no good just having an image if you don't know which patient it is or the body when it was taken which contrast agents have been given y ardi ardi ardi so basically what you have with DICOM is you have the pixel data and you have a whole load of other stuff that goes with it it's even got its own network protocol believe it or not because it was invented about the same time that HTTP was being written there's variations to these HTTP now but at the time nothing else existed so they wrote their own and it's a very unusual standard DICOM because it encodes everything from the semantics you know in terms of when I'm talking about XYZ it actually defines in patient coordinates which direction they shall call X which is actually left to right which shall be Y which is front to back and which is Z which is head to tail and they even define that for animals as well which really gets fun now I don't want bloody updates right I don't know how many of you at the previous talk but they had the same problem right so that's what this standard is it's actually quite fun to be involved in I'm one of the main troubles of it we actually had a 10 year birthday party for the standard a few years ago which is sad isn't it really I couldn't quite find the picture of the birthday cake but I got one somewhere anyway most imaging standards have got metadata I'm not expecting you to read all the details but this is a picture taken by my IT manager of the favourite little character he's got on top of his monitor in our office and it's interesting to see just how much JPEG data where there is the tin foil brigade might find this really interesting to be able to see the the coordinates etc but anyway DICOM takes this to a new level because here we have a CT again this is my father's as you'll see in a moment because the details are in there here's page one of the metadata page two including his name at the top page three this is where you've actually got some of the real fun details about the image itself like the image matrix 512x512 the real image size 0.437mm per pixel 16 bits one thing that's quite common about medical imaging is that most of it is done to more than 8 bit most is 10, 12 or 16 bit not sure of the colour most of that is still 8 bit but most of the CT etc was actually done to a higher bit depth and then a grayscale transformation called windowing is applied at the time but three others I've marked which are really really important when we come to look at the 3D which is the reason I've underlined them is we've got the image position which is relative to an arbitrary co-ordinate location but a defined co-ordinate directions so that tells you where that particular slice was taken and also its orientation which you'll notice is not actually 1 0 0 0 1 0 which is what you would expect if it was just along the X and Y planes it's actually slightly tipped which makes it real fun when you come to put these images together because they're often taken not like a pack of cards but like a pack of cards that you've slightly slipped sideways so you don't actually have a cuboid you've got a parallel gramoid which really makes the 3D reconstruction quite fun then all the rest of these attributes including one interesting one at the bottom the pixel data because as far as dichom is concerned the pixel data is just another attribute there's lots of attributes of these images and one of them happens to be the pixel data you can actually have objects without pixel data but that's getting rather silly so pretty much for the last 20 years every manufacturer of digital of medical imaging equipment in the world has complied with this dichom standard and they can all now send their images into the hospital's picture archiving and communication system which is a huge great repository I mean they're typically hundreds of terabytes some run into petabytes which use the defined archiving a defined indexing levels of a patient has a number of studies a study has a number of series and a series has a number of images and they're pretty much modality independent there are some dependencies in terms of the low level data for instance you know tube killer voltage which we saw there on the previous CT would make no sense for ultrasound and so on but the actual main bulk of it is pretty much independent so the same packs can handle CT MRI ultrasound and all the others which is great and then it makes the images available back out to workstations for people to look at right I made a joke when I talked to Johnty about well you know if I'm going to talk about this I could really try to send people to sleep with a talk about compression and he said no don't this is EMF you can't be too techy for EMF so if you do find this bit too techy and boring blame Johnty not me so there are various ways of compressing images you know you all use the fact that your JPEGs are far far smaller than raw images medical images are so huge there's a lot to be gained from compressing them not least of which of course is an emergency situations as I was doing for a home review if you want to be able to get them down limited bandwidth lines you can get them there a lot faster if they're compressed I won't go into the details of lossy versus lossless here there's actually a lot of work chain you can do lossless compression with no loss of information whatsoever all the techy people agree on that just can't get the lawyers to but certainly lossless compression the medical equivalent or imaging equivalent of ZIP is well worth while having so I'm just going to talk about that for a moment because it's quite interesting the fact that everybody here knows ZIP they know how it works I'm sure you all know it's looking for repeating patterns which is why it works marvelously on text and things where the letters T-H-E are going to keep on appearing in English and keep on getting replaced by smaller and smaller and smaller tokens as the system goes through you can't really do that with imaging because you don't have repeated patterns as such the predictability you've got in imaging is different it's the fact that most pixels in an image aren't that different from the pixels around them so it's the differences that you're looking at not the absolute values and not the absolute patterns so that's what's normally used I'll talk about it more in a second so that said for the video and things when you're wanting to do the ultrasound box standard MPEG's got a lot to be said for it and that can actually be encoded within DICOM DICOM allows the images to be uncompressed or losslessly compressed or lossy compressed or even for multi-frame images it can be MPEG compressed but still have all the surrounding metadata to explain what's going on so here we are, he's actually doing some lossless compression yes I know it's very techy this but what you do is you calculate what the predicted value from a pixel would be based on the pixels that have been decoded so far adjacent to it you then calculate the offset which in most cases is going to be 0 or 1 most pixels don't differ much from the pixels around them you then Huffman code the number of bits that are required to store that data followed by then providing the data itself so in this particular case I've used an example the Huffman coding by the way is arbitrary it's typical ish so 2 0 0 0 minus 1 7 would become coded as 2 comes into the range plus or minus 2 to 3 so it has that as its Huffman coded value followed by 2 bits which are 0 0 in this case the 0s only need 1 0 bit the minus 1 comes into the plus or minus 1 so it has Huffman coding of 0 1 in which case the value of 0 would mean plus 1 1 would mean minus 1 that's the 1 there and then you've got the 7 which is encoded further on so you've taken what would have been if this was 16 bit data 80 bits 5 values each time 16 bits and you've ended up encoding them up I haven't even counted it to be honest 8 I think it's about 17 bits so you've got about 3 to 1 compression there and that's about what you typically get with this lossless compression and the other thing is you do actually get back to exactly the image you started with as ever it remains like zip it relies on having predictable images apply this to noise and it will probably expand the image rather than subtract it but that's what you get getting images to workstations the Dicom protocol can be used but again this Dicom because it does everything from specifying the format of the images through to the network protocols to your own query language it's like a version of SQL almost because it predates widespread adoption of SQL this is a 25 year old standard but it does mean you can find all the images for a particular patient and call them back I'm going to jump through this bit a bit otherwise we're going to go on but you then have to look how are you going to review such a large quantity of data well believe it or not when I left clinical medicine 15 years ago we were still taking digital images out onto sheets of 17 inch by 14 inch film on fixed grayscale mappings and then reviewing them out on a series of monitors like this guy here absolutely crazy but that's what we were still doing 15 years ago digital to analog to look at which then loses all the flexibility you could have then the next step was to go digital but to try to replicate whatever you've been doing in the analog world there was this stage where people had banks of 21 inch CRT monitors which was not much fun when combined with the NHS rule they were not allowed to have air conditioning in offices imagine these going back 15 years in a small office and it gets very warm yes we could do this in here as well but at least we've got some air then people moved forward to looking at them in real time through smaller numbers of monitors then they realised that in fact with a modern computing power you could actually look at these images in different directions and I'll be showing this live in a moment but these images were obtained by taking slices this way out of then being reformatted into slices other ways and you can look through what's called maximum intensity projection and then you can actually do volume rendering if you really want to produce fancy and fun looking images I say that carefully because they are great fun to produce and I'll show you some now in a moment the medical usefulness of the volume rendering it's great for showing patience it's great for impressing at sales conferences in practice not many people actually do it but the big thing nowadays and this is what I'm going to talk about next in the remaining 5-10 minutes so I've got is that of course you don't need the supercomputers you used to need to be doing this your average GPU like what I've got here integrated graphics on my laptop is actually good enough to be able to be doing this live and in real time with really quite simplistic programming so if I just quit out of this for the moment I hope this is working when it moves over right so there we are this is the image I was looking at just now this is actually my father's brain again two images, the green and red lines anybody work up what they are sorry no, they are actually where each image is on the other image so if I scroll that one backwards and forwards you see the line on the other if I move this one which is moving side to side it moves that one and you can do all sorts of wonderful things that would make you sick if you did this with your real head you can just spin them around and so on and again all of this this is actually a single 3D texture held in the GPU and I'm just re-rendering that same texture again and again with a different matrix to say how to be applying it and that's it's relatively simple, this is simple cut plane so you only need to do one operation per pixel it's really quite trivial this is a bit more fun maximum intensity projection this is a fancy MRI I didn't talk about it earlier but one thing you can do with MRI is you can actually do more fancy things like you can look where the things are moving you can look where they're flowing and this is an angiogram this is looking at the blood vessels in the neck done with MRI without having to inject contrast medium without having to go putting a needle into somebody's groin and inject all the stuff in that you used to have to do but the thing about this is you can you can just do this you can look around one little nice oddity in view of what I was saying about phasing coding earlier can anybody work out what this bit here is go on no sorry no, it is chin you've encoded in that direction using phase if you go too far phase can only go from 0 to 360 degrees if you go too far it wraps and what we've actually ended up with because they didn't quite get the positioning right is that his chin which should be here has actually wrapped around to the back of his neck here and therefore I know looking as a radiologist looking at this picture I know which was in terms of what I was talking about earlier which is the y direction and which is the z direction the z direction, the phasing coding direction is the front to back because that's the only one that can wrap so we've got the wrap on there but one nice thing you can do with this suppose you wanted to look at things in detail again if I go into a straightforward axial view looking down I've got the ability again to just in a GPU I can do a cut and spin it round from there so you can move stuff out of the way that's blocking and again, this isn't a supercomputer this is a bulk standard business laptop with integrated graphics going back to my father's head again because it's a bit more fun this one there's a prize for anybody who can work out what was actually wrong with my father and why he had to have this done because there's a few slight clues on the top of the head nope nope no, there were times I feel like I would actually like to do that myself but no bag on I'll have to work out what the prize is you're welcome to one of my company mugs that's the prize I didn't warn you what the mug was but in fact the cutting for here is really great fun because if you want to look at the base of the brain which is really a lovely fun structure I can do that now if I spin you're looking down inside the brain and again, just pointing out this is all on that this is just using a GPU this technology which was it's been invented and optimised for games but boy is it wonderful for ideology and then I've got another image here this is a jaw it's a bit slower because this is a higher I think this is a thousand by a thousand by 500 or something there's basically a gigabyte of data here but in fact it's 64 bits 16 bit data and again, if you want to look from behind oh dear, I've got stuff in the way so if we go back to let's say the axial view I'll put a cut line on, a cut rectangle cut round here go back to rotate let's do it this way so I can see where I'm going and now you can see the back of the teeth nice and clearly none of this is rocket science but it does give me an excuse to go back into my last slide, which is just about right because I've got a couple of minutes left which is to say that I actually what I actually do is my business, I'm a sponsor here so I believe I'm allowed to talk about my business for a few minutes but unlike most sponsors, I'm not actually here to sell I very much doubt whether any of you guys here who actually want to buy a DICOM toolkit if you want to yes please, come and talk to me but what we do is we don't do the primary imaging we do all these sections I've labelled up here we provide a software toolkit which enables people to convert raw data into DICOM to handle it to do the network protocols the querying, the accessing and all that 3D rendering is actually our own software as well so what I would like to say obviously if that's what we do we do DICOM toolkits, we've got hundreds of customers around the world we're supporting other developers we don't have to deal with plebs we don't have to deal with the end users we only deal with other developers so it's a great fun job and guess what we could do with an extra person or two so if any of you find this particularly exciting if you're a hardcore programmer you'd love the challenge of working in a totally unstructured environment with multiple different languages come and talk to me or find me in the beard tent afterwards and I might have a job for you at that point I'm finishing thank you very much you've got about three minutes left so if you'd like to take two questions or so I'll take the two closest which are these two here thanks very much just a quick question about the MRI scanning that I might have missed you get frequency domain back from the receiver do you and is it just a single frequency pulse or do you actually vary the frequency of the transmitter as well I was talking in terms of putting a single frequency pulse in what you actually do is you put multiple frequency pulses in and you actually image several different slices at the same time and there are ways of doing it with minimal interference because of when the echoes are timed so certainly the generator can produce multiple different frequencies and there are uses for doing so but the simplistic version I was giving here what would be the one slice would be a single frequency in and then a multiple frequency bandwidth limited signal out two small questions first of all I recall that in radiology usually you use black and white grayscale monitors with a higher resolution like 10 bit is this still a thing or have current displays catch up and I like you also to say something about the amount of gain you can get from JPEG lossless compression which I just found out today that it exists never heard of it how much data can you actually save using that first question second question first because the easy one because it's data dependent the degree of compression you get depends on the data you put in a bit more precisely the entropy content you typically get about two and a half to three and a half to one compression so you're typically saving about sort of 60 to 70% but it does vary according to the noise level the noise of the data the less compression you get and it's actually defined in the same standard as defined the original JPEG believe it or not that same ITU specification defined the one defined the lossless it's really used outside DICOM the monitors there's a lot of I deliberately stayed out of it I had a job to squeeze everything there into my hour so there are all sorts of standards about how many bits output you need to be displaying and there's a consistency standard what more and more the work is showing in fact is that actually properly calibrated 8 bit will actually do you fine so what many people are now doing is taking 8 bit with calibration in the video card to produce the right look up table in the card rather than the previous approach which was calibration in software which then required a 10 bit output card but what most people completely forget is they do all their careful calibration to get it absolutely right in a darkened room and then they sit there actually trying to use their carefully calibrated monitor in a room with the windows open in which case your calibration goes completely to pot anyway and you might as well not have bothered more and more people actually getting FDA approval for viewing on iPads bit dodgy to me but they're managing to get it through the regulatory side I'm afraid that's all we have time for but thank you very much Dave and