[MUSIC] Okay, we'll talk about the non-Cartesian imaging in this video. So now, Cartesian imaging sequence is performed when we acquire data in a Fourier domain. So the acquired data points are directly stored in a Cartesian grid. So Fourier transform directly generates the imaging. This Cartesian imaging is performed in most clinical scanners these days too, okay. But there are many non-Cartesian imaging sequences. In this case, K-space trajectory is not on a Cartesian grid. So in case of Cartesian grid, the samples are shown like as shown here. So we just acquire data based on frequency encoding and the phase encoding, okay? That is conventional Cartesian imaging. But we can acquire data in a slightly different manner like as shown here in the spiral way or some radial way, okay? We can acquire data in so many different ways. So this K-space trajectory can be manipulated by modifying frequency encoding and phase encoding gradient. So as I mentioned when we talk about week four. The integration area of phase encoding and frequency encoding gradient determines location of that acquired point on the K-space, okay? So, because of that we can manipulate the location or trajectory of data location. So we can manipulate a spiral way or a radial way or a conventional Cartesian way, okay. They can be manipulated by just changing the gradient strength in combination with data location. So most anatomical contrast of an image is determined by low spatial frequency components like a central portion of the K-space. So because of that sampling more at the low spatial frequency offers a better image quality. So that is generally true. And in that sense, non-cartesian imaging has some advantage and it's one of the efficient ways to achieve this, to take advantage of this method, okay? So advantage of non-Cartesian imaging is that a faster acquisition is possible, okay? You can fill K-space with fewer excitation than the conventional Cartesian imaging, okay? So it can be generally used in cardiac MR imaging, okay? And it's more robust to motion. And K-space center is sampled every TR, okay, every excitation. So it provides some averaging effect across motion. There is no ghost artifact which occurs in EPI imaging. There is a lot of problem too, But compared to the EPI, so it's one of the efficient and fast aquisition way, but the artifact pattern is slightly different from EPI. So blurring may exist if we acquire data in a non-Cartesian domain. If the undersampling of the high spatial frequency component or some gradient imperfections, so this caused some blurring effect if they are not compensated well okay? They are not compensated well, okay? The reconstruction is not straightforward, okay? That's another disadvantage of non-Cartesian imaging. And that's why just the Cartesian imaging has been so popular, okay? Just applying a Fourier transform generates an image, okay? In case of non-Cartesian imagin, we cannot just directly apply 2D Fourier transform. We need spatial reconstruction algorithms to get images. Let's talk about one example of non-Cartesian imaging, the spiral imaging. So this is pulse sequence diagram. So after slice-selection and refocusing, we can apply gradient. We typically apply the phase encoding or readout prephasing in case of Cartesian. But in case of spiral imaging, it starts from K-space center. So we can directly start from data location, from right after the slice refocusing gradient is applied. And then we can keep acquiring data. During this data acquisition readout and phase encoding, both of them, can change in this waveform as shown here. And then the, Acquired trajectory is going to move like that, okay? So this is single trajectory and also this is example of multiple trajectories, okay? And then the K-space can be acquired using single or interleaved multiple trajectories, which is obvious. And the single trajectory is more time efficient, okay? This is one shot EPI, same as one shot EPI. And this can be considered as corresponds to multishot EPI, okay? It's similar. So single shot is more time efficient but it also is more susceptible to artifacts from field inhomogeneity. So definitely multishot improves image quality, much better than the single shot, but it takes longer time. K-space corners are sometimes ignored because the corners hardly contribute to the final image. Okay, this is example of radial K-space sampling. So excitation and slice refocusing is the same. But this phasing encoding, prephasing and readout prephasing is applied all together. So for instance, if we fill K-space line along this oblique direction then to move the position here, both readout and phase encoding and readout, prephasing gradient are necessary to move all the way here. And then we can read all the way from here to here and these gradient should be applied together. So depending on the angle of that K-space lines these gradient strengths can be manipulated, okay? This is concept of radial K-space sampling. So historically, radial K-space sampling was the first data acquisition method introduced in MR imaging. Like filter backprojection for the X-ray Computed Tomography CT, okay? So K-space center can be continuously updated because of this sampling scheme. K-space sample lines are continuously acquired. So this is desired to minimize the motion problem for the dynamic MR imaging like heart or lung, okay? And there may be some streak artifacts if this number of samples are not large enough, then there'll be some streak artifacts. Okay, and then how can we reconstruct a non-Cartesian K-space? I'm not going to talk details about that, but let me just briefly talk about introduce some representative methods. So one is using filtered backprojection algorithm which is used for the computed tomography image reconstruction. So for the radial acquisition that I just mentioned in the previous slide, the radial acquisition is almost the same as the CT, okay, acquisition. So that is also performed along multiple angles. X-ray imaging is performed along multiple angles for the CT imaging case. And similar thing is true for the MR imaging for the radial imaging acquisition. But that domain is on the frequency domain. The only difference is the MR acquisition is performed on the frequency domain but the CT acquisition is preformed on the spatial domain. So that is the difference. But we can use very similar algorithm called a filtered backprojection to reconstruct images. Or sometimes we can interpolate data to predict the data points on the Cartesian grid from the data acquired in the non-Cartesian grid, okay? And then we can use some interpolation algorithm to find the data points on the Cartesian grid. And then we can apply two-dimensional Fourier transform to get the images. That is another way to generate images from the non-Cartesian K-space data. Okay, this is going to be end of this course, and I thank you, all of you, to join this course. And I wish best luck in your future career. Thank you very much.
