 What I will be talking about are some advances in digital radiography, particularly in medical imaging. And I want this to be an interactive talk. So please feel free to ask questions and don't wait till the very end because I'm going to cover a lot of material. And I mean I don't expect all the material to be tractable to everybody but I am going to start with some fundamentals so we can start off at a point where some of the specialized material would be tractable to those people who are interested in that. Basically this is a very exciting field and digital radiography has come a long way and as you will see has got a myriad of different applications. So I mean in terms of the objectives of the talk, I will start off with some very simple fundamentals. They will be very intuitive and then we will talk about some design considerations of DR systems and some of it may involve knowledge of electronics but I will try to explain everything. And then we will talk a little bit about the applications of digital radiographic panels in I think called cone beam CT and then tomosynthesis which is an exciting new area in mammography. And then we will talk a little bit about medical image processing and the last one is computer aided assisted diagnosis and 3D visualization. The last one particularly is an area which there is ample opportunity to do a lot of good work in India. So typically a radiographic imaging change starts with an x-ray source and then you have a low energy absorbing filter because low energies are basically they just add to patient dose and they don't add to any image information. So you want to reduce patient dose so people use filters of aluminum typically and then there is a collimation and then there is we will talk about all these components in a little bit of time. There is a collimating an anti-scatter grid and then there is a detector. I mean typically people are used to screen film systems in radiology which are the traditional systems where you would have a screen that a scintillator that converted x-ray photons to light photons which then went and exposed film. Digital radiography is a replacement to that and much more. Now we all know what x-rays are basically if you look at the electromagnetic spectrum basically x-rays are somewhere in the 10 raise to minus 10 meter wavelengths. And of course gamma rays are also there but we won't talk about them. They're used in nuclear medicine imaging in gamma cameras and also in gamma knives in radiation therapy. Now if you look at production of x-rays typically you have a cathode, you have a filament where electrons are focused onto an anode to a small spot. Sometimes the spot in mammography can be as small as 100 microns and then typically have an anode that's rotating because you don't want it to be heated. And so what happens is you have electrons impinging on the anode and creating x-rays. Typically only 2% of the energy that impinges on the anode is used to create x-rays. The rest of it goes away as heat. It's a very inefficient process. And of course the anode is also cooled. Typically there's an oil cooler and there's a bearing. And this whole area of x-ray tubes is a specialized area and there are just a few companies in the United States that make very special tubes. And if you take tubes that are used in computed tomography basically the currents can be as high as 2,000 milliamps. But in standard chest x-ray it's just in the tens or maybe 100 milliamps maximum. So essentially what happens is electrons, you know, kinetic energy is transferred to the anode as we spoke and some lots of energy is lost to heat and some are released as x-ray photons. So different wavelengths of photons would depict different energies. Now if you look at a typical x-ray spectrum, typically if you look at continuous, you see a continuous spectrum and sometimes you get what's called a k-edge which is due to the k-shell allotrons and that's good because after that you see an increase in energies. So this is an x-ray spectrum. This actually is a 25 kV spectrum. This I suspect is a mammography application. And so when we move along, actually there are a myriad of different radiological units, some old, some new. And if you look at these units, the activities of units are known in curies, but then the SI unit of activities in becquerels. And then we, rent gains is exposure. That's basically 10 raised to 2.58 multiplied by 10 raised to minus 4 coulombs per kilogram. That's photons in air. That's the ionization. And then rad is a dose. Basically what happens is it's a unit of dose, but then if you go down, rem is the unit of human dose because not all x-ray dose, not all energies are absorbed. Only some are absorbed. And then seaward is what we use these days. Most of the SI units like gray seawards and becquerels are the ones that are used today. So there's a combination of different units. So if you see people using different units in different countries can be quite confusing. We actually have tables that help us convert. So now what happens with interaction of x-rays and matter? You know you take some matter here and then let's call this the input intensity and this is the output intensity. And what happens is you have photons that are traveling at the speed of light. They impinge on the tissue. Some photons are absorbed and some pass through. It's mostly the photoelectric effect that's important in this situation. So obviously you have less photons coming out of the tissue than that go in. Whereas we'd have a perpetual motion machine. And if you look at the ratio of the number of photons coming out to the number that went in, we'd have an idea of tissue attenuation. Talk a little more about that. So essentially there is a thing that's called a linear attenuation coefficient that is used in x-rays. So you have an anatomy that removes the photons from the beam. And so you have the interaction as we talked about the diagnostic x-ray photons with the tissue due to the photoelectric effect. So attenuation is absorption and also scatter. Some of it is absorbed and some of it is scattered. We'll talk a little more about scatter. So if you look at n, the incident photons and the thickness of the attenuating material is t, then essentially the relationship between n sub 0 and n is an exponential relationship that is related to the thickness as well as the attenuation coefficient. Well, it's called a linear attenuation coefficient because it was linear in the log domain. Okay, yeah, I know. Yeah, it's nomenclature because people used to look at densities on film. And densities are the log of intensity. So essentially when you looked at film, it looked like linear attenuation and that is the reason for it being called a linear attenuation coefficient. It's a good question. So essentially if you deconstruct the linear attenuation coefficient, it is a multiplication of the mass attenuation coefficient multiplied by the density. So if you have higher density material, obviously, depending on the mass attenuation coefficient, you'd have a higher linear coefficient. But if you look at mu sub m for tissue, bone, fat and muscle, you see different numbers. Okay, so essentially if you take the z effective for bone and soft tissue, that's the nature or the reason for tissue contrast. But many times you cannot visualize soft anatomy and so you have to look at different energies. So let's take a look at intuitively what a radiographic image is. So differences in tissues causes different absorption of photons. Let's say a constant number of photons per unit area impinge on tissue. So in one case you have 1000 photons in both cases. One case you get 750. In the other case you get 250 onto the detector. And let's assume that this is a detector that can count the photons. And so the relative contrast is basically 750 divided by the total number of photons minus 250 divided by the total number of photons which is 0.5. But let's see what happens when there's a little bit of scatter because this is not like a laser beam. Basically things get scattered in different dimensions. So essentially photons that come out of here, some of them go to this part of the detector. And some of this, these photons go to the other detector. Now if you don't do anything about the scatter, you suddenly lost your contrast. And so what do we do about scatter? So there's a thing called a grid that is placed over here which actually impedes these photons that are coming at an angle depending on the shape and size of the grid. And so only the photons that are coming straight through are caught at the detector. But however we have improved contrast but we've lost photons. Now normally this would not be an issue but however as we'll see a little later noise is going to come into picture and that is going to affect contrast. So let's recap a little bit and also you don't want to expose a patient too much unless it's a therapy situation where you selectively expose selective tissue. So exposure is proportional to energy times the number of photons and dose is proportional to the amount of energy absorbed in tissue. Depending on the radiation type we can have high exposure and no dose. I mean you have neutrinos that are going through you all the time but I mean there's no dose. So x-rays typically are ionizing radiation and they do damaged tissue in different doses and different energies. So unlimited amounts of ionizing radiations therefore cannot be applied to the human body and limiting exposure obviously is going to limit the number of photons. So in a noise-free situation we may not need many photons however all physical phenomena are subject to noise and they're different noise sources. So in the next few slides let's look at what intuitively we have in terms of noise. So if you have n photons essentially typically if you take a Poisson process you get square root of n is the mean square number of what is the mean square of the noise in terms of the photons. So let's say I have 100 photons so I have 100 plus or minus 10 intuitively and then so because of this uncertainty I mean if you have a or b in one case you could have 50 plus or minus 5 photons in another case you'd have 55 plus or minus 5 photons but depending on the nature of the detector these things a may look more contrasty than b and vice versa under different circumstances. So noise is important in terms of discerning contrast between adjacent objects and pixels. So what do we do? We increase the number of photons and suddenly b is darker than a under all conditions of photon noise assuming that you have a totally noiseless system elsewhere because you can also have electronic noise. This is what's called quantum limited. We will talk more about it in detail because as we go past these intuitive concepts things are going to get a little more mathematical and I know you know I tried to simplify it but we can't avoid it. So what's the purpose of a radiographic detector? So it's to collect information related to the number of photons on each pixel after extra photons pass through a patient or a tissue sample. So this information generates a radiographic image that can then be displayed and manipulated. Traditional ways of looking at radiographic images were screen film systems where you expose the film with light photons converted from extra photons. That's why there's a screen which is typically a gadolinium oxysulfide scintillator that has an efficiency of about 40% or so that converts extra photons to light photons and then these light photons are in contact with film and then you get a radiographic image and that's the screen film system. Today in many areas screen film systems are actually being surpassed by digital detectors. Of course digital detectors still tend to be quite expensive. So in digital detectors ideally you should be able to count extra photons. There are such detectors that do count photons that use photons coming into pressurized gas and then you count the photons you could look at the peaks of the energies but those are not very common these days. I mean that's still in the research phase and there are a couple of companies that are dealing with these things and then you have conversion of extra photons directly into electronic charge. It's called direct conversion. Actually it's a little bit, the terminology here is sometimes confusing because people take digital radiography as direct radiography but digital radiography also has indirect conversion in which case the photons are converted to light and then to electronic charge and we will look at the advantages and disadvantages of both of these in the next few slides. So everybody knows digital images are 2D array of pixel elements. Now depending on the radiographic exam in a chest exam each of these pixel elements could be about 127 microns to about 200 microns and that's also true in fluoroscopy but when you do mammography you're looking at smaller pixel elements anywhere from 50 to 80 microns or sometimes even 100 microns and this is related to the size of the objects that you want to detect because if you look at mammography you're looking for microcalcifications that could be about 50 microns or so and as we'll see later that you have to have the appropriate spatial frequencies in order to visualize objects in different exams. So we talked about different conversions direct conversions typically the way was to use selenium detector I mean selenium had been well studied but this is very pure selenium because selenium was used in xerography as a photo conductor and that was originally used and it's still used and we'll talk about the advantages and disadvantages of that to convert x-ray photons to electrons and then indirect conversion uses intensifying screens to convert x-ray photons to light and then to electrons. Now intensifying screens come in different shapes and sizes see a typical intensifying screen such as a gadolinium oxysulfide screen in a screen film system also disperses light so depending on the thickness of the screen the spread of light so you would lose spatial resolution but there are newer type of intensifying screens or scintillators as we call them that are built in a crystalline way and the most common is cadmium sulfide which is activated by thallium so it actually converts light at room temperature so let's look at the basic principles typically this is direct radiography where you have a photo conductor and what happens here is when x-ray photons impinge on a photo conductor and typically you have for a 500 micron thickness of photo conductor it's an amorphous photo conductor you have about 5000 volts across it and we will talk a little more about that and so what happens is when x-ray photons hit the photo conductor you get electron and hole pairs so you collect them on a capacitor then you have a switch and very similarly in an intensifying screen you have an intensifying screen and you have a photo detector which has associated with it capacitors now this is typically if you look at the displays that you use the LCD displays you select rows and then you select columns and then you energize the columns this is the opposite of that you have rows and you have columns here as well and we will talk more about it as we go along so we talked about indirect radiography where you have a scintillator photo detector and then you have an electronic charge and let's look at this is a slide from GE this is an older slide actually things have improved this is typically the microstructure of a digital radiography detector you have a scintillator which is cesium iodide then you have an amorphous silicon array basically people use amorphous silicon transistors amorphous silicon photo detectors because these can be laid out relatively inexpensively on glass in large areas and then you have row and column, you have row selects and column readouts and then it's typically on a 70 micron glass substrate this is a very special glass that's made by Corning in upstate New York and this glass has properties so it doesn't fluoresce with x-rays in the areas of interest because that fluorescence can add to noise so it's very special glass and it's very, very flat glass and these are very fragile detectors if you drop one of these detectors you have kissed about $65,000 goodbye or more they can be up to 35 by 43 centimeters or 14 by 17 inches for chest radiography and some of them come in 43 by 43 centimeters these are large arrays and each of these detector arrays is very expensive and the people who manufacture them are people in China and some in Korea the people who manufacture displays are the ones that are contracted to make digital radiography panels so they take half a day off of production and make all the digital radiography panels of one company because it's expensive and right now it's small volume typically that will change as we see and we saw direct radiography where it generates electronic charge directly now this is a cross section of an amorphous selenium digital radiography panel and I'm going to spend some time here in terms of talking about this and we can have some questions as well from people who are interested what happens is as I said there's a high voltage power supply and you have typically 100 volts or so no 10 volts or so for each micron and so a 500 micron panel has about 5 kV so there's a high voltage supply that is biasing this particular structure there's a top electrode and there's a dielectric layer to separate the high voltage it's like a large capacitor and you have electron hole pairs the electrons go to the top they sit there and then the holes come down and they're deposited into these capacitors these capacitors are typically one to two picofarads and we convert the charge in these one to two picofarad capacitors to a resolution of 14 to 16 bits and it's quite a feat when you take 2,000 of these things down a column and you're able to do it and so the electronics is quite sophisticated and then these are amorphous silicon thin film transistors actually the on to off ratio of each of these transistors is about 10 raise to 8 the on resistance of each of these transistors is about 2 to 5 mega ohms so this is typically a amorphous selenium structure and people used to think that this was the best but what happens as in amorphous selenium as is the case with any amorphous material the conduction happens between the valence band and the conduction band so there are states in between and there's what's called hopping conduction where within the band the charge carriers hop from one end to the other and if you look at the mobility of a amorphous selenium or amorphous silicon it's 1,000 times less than that of standard crystalline silicon and then because this is such a large piece of material you have what are called traps and some of them are called deep traps so as the electrons start moving and then there's a thing called the mobility recombination time product you want to make sure that it's large enough that these things don't recombine while they're moving some of these things can go into deep traps and when they go into deep traps not only do they create space charge they also create, they just sit there and you can get them out and in order to get the deep charge, deep traps out what is done with some of the amorphous selenium structures is that it's flooded with light but this also causes a phenomenon called ghosting where that means history of the previous image stays with the structure even for the next image depending on how it was exposed and that is a problem for high speed radiography which sometimes you use in fluoroscopy and actually today as we will see and I will go over that in a moment cesium iodide with crystalline structures are the preferred way of doing things because that has come a long way now if you look at the collection of charge you have a switching control that selects rows and then you have columns and you have pre-amplifiers and analog to digital conversion and then it goes into a digital processor and then you have image display and this is an early digital radiography detector this is from a company called NRAT that in Canada that's owned by analogic this is one of the early selenium detectors and they use tab bonds to bond all the electronics today the better ways of doing it early five years ago you can see this is the detector it's huge and so let's look at what happens intuitively in terms of charge collection so think of the pixels as buckets really these are buckets of electrons and so you select the rows let's say we selected the first row then each of these rows are connected to the columns then that charge actually comes down and collected in charge amplifiers so essentially this is a common bucket which each time you select rows you could have anywhere from 2000 to 3000 rows and so you have to worry about interaction between each of these pixels and that is why the on to off ratio of the amorphous silicon transistors the TFTs has to be 10 raised to 8 because you need to maintain about 14 bits to 15 bits of isolation so this is typically and we will go over this a little more I mean for those of you who are interested in how the electronics is done now if you look at this this is just an equivalent circuit so you have a photodiode here and you have the storage capacitor and this is the gate line and this is the integrator basically the other things over here which we will talk about pass the charge amplifier but typically what happens is we have a virtual ground here so what happens is the effects of the column capacitance are negated by the gain of the loop gain of the amplifier see column capacitance can be as high as 50 picofarads which is fairly high but it does come in in terms of the electronic noise but you don't want the charge that is in the pixel to be trapped in the column capacitance so there is an integrating capacitor here that is typically 0.5 to 1 picofarads and then you have a switch here to basically reset it and then you get a voltage here and typically the way it's done is there are ICs which have 128 to 256 charge amplifiers they cater to 128 to 256 columns and some of them have A to D converters inside them and some of them have A to D converters on the outside so this is highly integrated and basically you have a whole bunch of charge amplifiers and it's a fairly specialized technology so there are different ways of doing charge amplifiers and actually General Electric uses a bunch of delta sigma converters for digitization but prior to each of these there is a charge amplifier so we go a little more into details now if you look at this this is a pixel and there is a switch these are some of the traditional panels and now you notice here that this pixel is not occupying the complete area you know you need space for the rows and columns and you also need space and this is the second generation version in the third generation actually we put the electrics underneath the pixel but you still have the second generation panels where this space is taken away and this is a constant amount of space so if a pixel is small a greater percentage of space is taken away so this is what's called a fill factor and it affects the efficiency in terms of collection so now we are coming back to our friend Scatter so you have scattering of photons occurs in many places you have X-ray photons that scatter at the patient then you have light photons that can scatter in some scintillators we talked about the standard scintillators where they could be scattered so scatter as we know causes the contrast of the image to be blurred and it can be reduced in certain scintillators like CSI that are grown as crystals so it's almost like they have fibers inside them so the light goes down a pipe the high voltage that is applied to photo conductors prevents electrons from scattering that's one thing it does that's why photo conductors were originally before we started seeing problems in terms of deep traps where things of choice so this is this is an older slide and basically I look at direct imaging and scintillator photo diode arrays to be very similar but this was generated by a company that was selling selenium photo detectors and I could come up with a better slide and therefore I use this slide but if you look at but let's look at the profiles if the signal profile and screen film systems basically is a little blurred and in computer radiography which is another area of early digital it's still quite popular techniques and was invented at Kodak in 1975 but what happens is instead of a screen film you have barium fluorobromide that Europium activated as a sensor so what happens in computer radiography is that electrons that are generated in the scintillator due to the x-rays are trapped in metastable states so then you take a laser and read it point by point it's a red laser so the trapped electrons are sitting between the conduction and the valence band so they move up to the conduction band and fall down so there's more energy that comes out so it generates blue light and people use photo multipliers to look at this light in terms of a point by point scanning and it's still a one billion dollar industry one of the advantages of computer radiography was that was truly retrofitable you take a screen film cassette and take a computer radiography cassette in a standard screen x-ray generator bucky combination bucky by the way is a term for a screen grid that moves back and forth so that you don't see the grid lines then some of the earlier indirect imaging you'll see ccd detectors with scintillation screens now I won't go into details but it depends on the minification and magnification factors in terms of the optics the optical transfer function convolves with the modulation transfer function of the rest of the system and you get a signal profile that's this wide and one of the early companies that actually sold quite a few systems was Swiss ray now today this is very common you have a scintillator which is about a 500 micron scintillator or 150 micron scintillator when it comes to mammography and then you have a photodiode array and in direct imaging you have anywhere from 500 microns to 1000 micron material of amorphous selenium it tends to be quite heavy and these are the different actually if you go from the screen film to the cesium iodide you see an evolution in terms of improvement in terms of the sharpness of the image and also the efficiency so let's look at some other requirements for detectors in general radiography you could have detectors greater than 40 centimeters by 40 centimeters typically today the standard is 43 by 43 or if you want to do chest radiography it is 43 by 35 centimeters corresponding to 14 by 17 inches then mammography you have 18 centimeters by 24 centimeters and if you look at the pixel sizes in general radiography you can go from 100 to 200 microns mammography 60 to 100 microns and the energy range is 30 to 120 keV the low end usually are used in pediatric exams and mammography it's around 20 keV okay in terms of detector properties you have the coverage of the field you have geometrical characteristics and your quantum efficiency we'll be talking about all these things you have sensitivity of the detector you have spatial resolution you have the noise characteristics dynamic range uniformity okay acquisition speed frame rate, frame rate is important when you are doing fluoroscopy or dual energy imaging and cost I mean typically today's detectors are very expensive but they will come down in the future with different technologies and we will touch upon those in this workshop detector properties we talked about it the tiling which is the size of the tile of the detector and then the fill factor which we talked about basically how much of the detector is basically looking at X-ray photons and how much of it is being used for extraneous things necessary but extraneous such as the electronics, the tin firm transistors this is tiling basically I mean it shows a rectangular tile but usually it's a square tile then if you look at an object a detector pixel element will always distort the object and it's going to be a transformed object actually it convols with the transfer function of this and so you will typically get an object that is slightly larger than the object that you are looking at depending on what the transformation characteristics are fill factor is what we talked about this is the active detector area and this is the area that is therefore the electronics and also the lines that you have in terms of rows and columns then spatial resolution you have things called outside the system factors which the effective size of the focal spot now if your focal spot is fairly large you cannot do any better than your focal spot because there is already a transfer function there that blurs the image and that's why you try to use smaller spots for exams like mammography magnification when you magnify things the spot also gets magnified and then relative motion I mean unless you are looking a cadaver or a dead body I mean there is always motion in mammography of course people have compression plates where the object to be imaged is held constantly but that's not true in imaging and so typically what one does in terms of x-ray imaging is you try to use a small pulse of x-ray with a higher current so it's like a shutter in the camera well they could be anywhere from 8.3 milliseconds wide to about anywhere to a second depending on the exam now if it was in an extremity like a hand where you can hold it fairly steady the exposure is talked about in milliamps multiplied by seconds so you have MAS whereas in CT scanning where you are going very fast use very small pulses but you use very high currents like about 2 amps so now in terms of detector characteristics you have the aperture size of the detector you have spatial sampling and signal spreading effects aperture is the active portion of the detector element so it also determines the spatial frequency response and we will talk about that in a moment and we will talk about a lot in a few moments and the sampling interval we will talk about spatial sampling you have a Nyquist frequency and aliasing will occur let's talk about intuitive sampling and aliasing let's say you have two objects here one has got an intensity of other has an intensity of minus 1 and the detector element is twice as large as this so you get a 0 I mean that is essentially a Nyquist in terms of spatial sampling whereas and the same thing happens if you have plus 1 and minus 1 whereas if you have an element and a pixel element that are compatible this plus 1 gets transformed to plus 1 so yeah well radiation does behave like light before it strikes any object but after that it scatters of course it is like light scattering too no it is not possible to focus it people have tried focusing there are very specialized ways of focusing it people have used very thin capillaries of material there is a Russian patent on that and actually some people in University of Syracuse were doing it but it was for such small doses the focusing part of it takes away a lot of energy so it makes it very very inefficient so it cannot be focused like light it can be unfortunately so what we do is we convert it to light and then if necessary focus the light it is what is done in CCD type of detectors now usually if you look at some of the detector have I answered your question are people have done that actually is done in multiple in in multiple energy imaging so some of the energies pass through the sensor on top and some of them pass through the sensor on bottom and so different energies of the sensor are used to image the patient what happens is contrast is also energy dependent it is also dependent on the spectra so those are some of the things that are done but what happens in typically digital radiography detectors it is expensive enough to make one array on glass and in future actually to answer your question people are looking at putting amorphous silicon on plastic University of Waterloo in Canada they are doing a lot of work what happens is when you deposit amorphous silicon on glass people had a 250 degree Celsius process and that could not be used for plastic because it would melt the plastic but today there are low temperature processes that go at 150 degrees Celsius on polyamide substrate that are going to come into the picture a few years from now and there is some work that is going on in every place and because of the confidential nature of this work you do not hear about it unless it is within a particular company now essentially if you look at distribution of interacting quanta it is typically follows a POSOP statistics and I have showed a reference there so you have sigma as n sub zero which is the mean of number of x-ray quanta following on a detector of a given area and then you have the probability of success basically is this a useful quantum or not so if you have g as the mean gain then the signal can be described as n sub zero multiplied by g multiplied by the probability of success and the variance is given in terms of an extra term in parentheses that is gamma square plus sigma square and you can look it up actually in the reference will not go into details here but these are derivations that have been done by Harry Barrett and swindle many years ago so what you need to do is you need to incorporate all the sources that contribute to the noise of the signal and if you look at independent statistics so the noise will add as the square of the sum of the squares so the sum of the square noise is the sum of the squares and then you have to consider the spatial frequency dependence of the signal and the noise in order to completely define the noise because the noise is also frequency dependent noise is usually described as a noise power spectrum or the venous spectrum and the signal we have a thing called a modulation transfer function a lot of these terms have been borrowed from optics and then we have to correct for non-linearities and then typically we try to use dqe as a function of frequency which is related to the signal to noise ratio out as a function of frequency so what is this dqe detective quantum efficiency is the signal to is the ratio of the signal to noise ratio square of the output of a system divided by the signal to noise ratio square of the input of the system