Transcript Slide 1
2D FT Review MP/BME 574 1D to 2D Sampling • Signal under analysis is periodic • Signal is ‘essentially bandlimited’ • Sampling rate is high enough to satisfy Nyquist criterion • Other assumptions (for convenience) – Signal is sampled with uniformly spaced intervals 2D Sampling/Discrete-Space Signal N1 N1 x(n1, n 2) where n1 2 2 N2 N2 n2 2 2 n2 n1 2D Functions • • • • Impulses Step Sequences Separable Sequences Periodic Sequences Line Impulse n1 0 1 T (n1 ) 0 otherwise n2 n1 2D step n1 0 1 uT (n1 ) 0 otherwise n2 n1 2D step n2 0 1 uT (n2 ) 0 otherwise n2 n1 2D Step n2 1 n1 , n2 0 u(n1 , n2 ) 0 otherwise n1 Separable Sequences 1 n1 , n2 0 u (n1 , n2 ) 0 otherwise n2 Separableif , x(n1 , n2 ) f (n1 ) g (n2 ) e.g. u (n1 , n2 ) u (n1 )u (n2 ) n1 Periodic Sequences x(n1 , n2 ) x(n1 N1 , n2 ) n2 x(n1 , n2 N 2 ) for all (n1 , n2 ) periodic with period N1 N 2 n1 2D Convolution y(n1 , n2 ) x(k , k )h(n k , n k 1 2 1 1 2 k2 ) h(k1,k2) x(k1,k2) k2 k2 (3) (1) k1 (4) (2) k1 2D Convolution y(n1 , n2 ) x(k , k )h(n k , n k 1 2 1 1 2 k2 ) h(-k1,-k2) x(k1,k2) k2 k2 (2) k1 (4) (1) (3) k1 2D Convolution y(n1 , n2 ) x(k , k )h(n k , n k 1 2 1 1 2 k2 ) h(4-k1,3-k2) x(k1,k2) k2 k2 3-k2 (2) (4) (1) (3) k1 4-k1 k1 2D Convolution y(n1 , n2 ) x(k , k )h(n k , n k 1 2 1 2 k2 ) h(n1-k1,n2-k2) x(k1,k2) k2 1 k2 (2) (2) (1) (4) (4) (1) (3) (3) n2 k1 n1-1 k1 2D Convolution y(n1 , n2 ) x(k , k )h(n k , n k 1 2 1 (3) 2 k2 ) h(n1-k1,n2-k2) y(n1,n2) n2 1 k2 (2) (1) (7) (7) (4) (4) (4) (6) (4) (6) (1) (2) n 1 (3) (3) (3) n2 n1-1 k1 2D Convolution y(n1 , n2 ) x(k , k )h(n k , n k 1 2 1 (3) 2 k2 ) h(n1-k1,n2-k2) y(n1,n2) n2 1 k2 (2) (1) (7) (7) (4) (4) (4) (10) (6) (4) (6) (1) (2) n 1 (3) (3) (3) n2 n1-1 k1 2DFT Imaging in MRI MP/BME 574 Abbe’s Theory of Image Formation From Meyer-Arendt No Magnetic Field = Random Orientation No Net Magnetization Dipole Moments from Entire Sample Magnetic Field (B0) Magnetic Field (B0) m m Positive Orientation Negative Orientation Precession TorqueN m Bo L Bo Precession and Electromotive Force (emf) or Voltage • emf derives from Faraday’s law – Time-dependent magnetic flux through a coil of wire – Induces current flow – Proportional to the magnetic field strength and the frequency of the field oscillation d em f (t ) V (t ) dt B dS Coil Area Example z B1(t) y x B1 (t ) B1 sin t xˆ B1 cost yˆ Coil Area V (t ) L/2 B1 sin t xˆ B1 cost yˆ dS B sin t d d sin t 2 B L BL2 cost dt dt L/2 dydz xˆ L / 2 L / 2 Example z B1(t) y x B1 (t ) B1 sin t xˆ B1 cost yˆ Coil Area V (t ) L/2 B1 sin t xˆ B1 cost yˆ dS B cost d d cost 2 B L BL2 sin t dt dt L/2 dxdz yˆ L / 2 L / 2 Complex Voltage/Signal: General Case Vo cos( t ) Re(Vo e j ( t ) ) V (t )coil1 Re(BL2 cos t iBL2 sin t ) V (t )coil2 Im(BL2 cos t iBL2 sin t ) V (t ) 2 BL2 ei t rf-excitation By reciprocity, d 3 V (t ) B ( r ) M ( r , t ) d r 1 dt Sample S (r , t ) o t 3 M ( r , t ) sin( B t ) dt B ( r ) sin( t i ( r )) B ( r ) cos( t i ( r )) d r xy 1 1x o o 1y o o Sample 0 Rotating Frame Lab Frame After Haacke, 1999 Quadrature Conversion in MRI (and Ultrasound) Signal Processing Received Radio Frequency Echo Signal x(t) (fc = 10MHz; 40MS/s) X LPF xc(t) I—Channel 2 cos ct -p/2 Phase Shift -2 sin ct X Q—Channel xs(t) LPF In a high-end ultrasound/MR imaging system this conversion is done in the digital domain. In a lower-end system the conversion is done in the analog domain. Why? Spatial Encoding Torque N m B1 o Bo fL Bo 2 Consider m ore general B(r ) [ Bo G r ]k Bz Bz Bz ˆj where G iˆ kˆ x y z Slice Selection Ideal, non-selective rf: S(t) =rect(t/Dt) B1ideal(t) B1ideal Bxye i ( ot o ( r )) S non (t ) Bxy (r )e Bxy (r )e i ( ot o ( r )) i ( ot o ( r )) rect (t / Dt ) rect (t / Dt )dt Bxy (r ) Dt Non-selective rf-pulse Entire Volume Excited FTdemo: Rect modulated Cosine FTdemo: Rect modulated Cosine Spatial Encoding Gradients z B(r) r y x Consider more general B(r ) [ Bo G r ]kˆ Bz Bz Bz ˆj where G iˆ kˆ x y z Slice Selection Selective rf: Ssel(t) = sinc(t/t) rect(t/Dt) B1ideal(t) Apply spatial gradient simultaneous to rf-pulse. Df f ( zmax ) f ( zmin ) ( Bo Gz zmax ) ( Bo Gz zmin ) 2 2 Gz Dz 2 1 f Df / 2 rect ( f / Df ) 0 otherwise S sel (r , t ) Bxy (r )e i (ot ( r )) rect (t / Dt ) Df sinc (Dft ) 1 t Slice Selective rf-pulse Slice of width Dz Excited FTdemo: Cosine modulated Sinc Summary • Spin ½ nuclei will precess in a magnetic field Bo • Excite and receive signal with coils (antennae) by Faraday’s Law • Complex representation of real signals – Quadrature detection • Reciprocity • Spatial magnetic field gradients – Bandwidth of precessing “spins” • Non-Selective rf pulses using Fourier transform principles – Shift theorem etc… applies Spatial Encoding Gradients z B(r) r y x Consider more general B(r ) [ Bo G r ]kˆ Bz Bz Bz ˆj where G iˆ kˆ x y z Frequency Encoding f ( x) ( Bo Gx x) 2 Df Gx FOVx 2 f, B Df B=Bo FOVx xmin xmax For com m on bandwidthsGx 1Gauss/ cm Frequency Encoding Recall Lab 2, Problem 4: Piano Keyboard Time varying signal 2 1 0 -1 -2 0 0.002 0.004 0.006 Time (s) 0.008 0.01 0.012 80 60 40 20 0 -1000 -800 -600 -400 … A, 220 Hz Middle C -200 0 200 Temporal frequency (Hz) 400 600 800 1000 … E, 660 Hz Frequency Encoding Time (t) f ( x) ( Bo Gx x) 2 Df Gx FOVx 2 FT Proportionality Temporal Frequency (f) Position (x) Frequency Encoding sdet (t ) e iBot M xy ( x) e iGx xt dx s (t ) M xy ( x) e iG x xt dx Gx iG xt M xy ( x) s (t ) e dt 2 iff Gx t is equalto the conjugateof positionx. x Frequency Encoding Spatial Frequency (k) Time (t) In general, Proportionality FT FT t k x (t ) Gx ( s )ds Proportionality Temporal Frequency (f) k x Gx t 0 Position (x) Phase Encoding y s (t ) M xy ( x) e f, B Df B=Bo FOVx xmin xmax iG x xt dx dy Phase Encoding y s (t ) M xy ( x) e f, B Df B=Bo FOVx xmin xmax iG x xt dx dy Phase Encoding y f, B Df s(t ) M xy ( x) e iGx xt i ( y ) s(t ) M xy ( x) e iGx xt iG y yT B=Bo FOVx xmin xmax e e dx dy dx dy Phase Encoding B y Zero gradient for time, T y Phase Encoding B y Positive gradient for time, T y Phase Encoding B y Positive gradient for time, T y Frequency Encoding Spatial Frequency (k-ko) Time (t) In general, Proportionality FT FT T k y (t ) G y ( s )ds Proportionality Temporal Frequency (f) k y G yT Position (x) e-iGyT 0 2D Fast GRE Imaging TE Gx Phase Encode Asymmetric Readout Gy Dephasing/ Rewinder Gz ShinnarLaRoux RF RF TR = 6.6 msec Dephasing/ Rewinder 2D FT t ky Finish k x (t ) Gx ( s )ds 0 T n kx Start k y (t ) G y ( s )ds 0 G yT nDky, where Ny 2 n Ny 2 3D Fast GRE Imaging TE Gx Phase Encode Asymmetric Readout Gy Phase Encode Gz Dephasin g/ Rewinder ShinnarLaRoux RF RF TR = 6.6 msec Dephasing/ Rewinder 3D FT kz Tscan =Ny Nz TR NEX i.e. Time consuming! ky n kx Summary • Frequency encoding – Bandwidth of precessing frequencies • Phase – Incremental phase in image space • Implies shift in k-space • Entirely separable – 1D column-wise FFT – 1D row-wise FFT Navigating in 2D k-Space • Goals – Improve your intuition • Specific examples – Effects of: • Apodization windowing • “Zero-Padding” or Sinc interpolation – Vendors refer to this as “ZIP” • Sampling the corners of k-Space Elliptical Centric View Order High Detail Information kz ky Sampled Points Overall Image Contrast MRI: Image Acquisition K-space FT Image space FT FT Case I kz ky Case II Case III Case I k-space: Image Space: kz DFT ky Bernstein MA, Fain SB, and Riederer SJ, JMRI 14: 270-280 (2001) Case II k-space: Image Space: kz FT ky Case III k-space: Image Space: kz FT ky a b Zero-padding/Sinc Interpolation • Recall that the sampling theorem – Restoration of a compactly supported (bandlimited) function – Equivalent to convolution of the sampled points with a sinc function Recovering or “Restoring” f(x) from f(n): 2s’ 1 Dx F ( s) F ( s) per rect( s 2 s ') sin(2s' x) f ( x) F ( s) f (n) 2s' 2s' x 1 Recovering or “restoring” f(x) from f(n): where N n' N sin( 2s ' x ) 2s' 2s ' x f(n) f(x) 2 Dx F ( s) F ( s) per rect( s 2 s ') sin(2s' x) f ( x) F ( s) f (n) 2s' 2s' x 1 2 Recovering or “Restoring” f(x) from f(n): 2s’’ 1 Dx F ( s) F ( s) per rect( s 2 s ') sin(2s' x) f ( x) F ( s) f (n) 2s' 2s' x 1 Recovering or “restoring” f(x) from f(n): sin( 2s ' ' x) 2s' ' 2s ' ' x f(n) f(x) Dx F ( s) F ( s) per rect( s 2 s '') sin(2s' ' x) f ( x) F ( s) f (n) 2s' ' 2s' ' x 1 Recovering or “restoring” f(x) from f(n): f(n’) where N n' N Dx F ( s) F ( s) per rect( s 2 s '') sin(2s' ' x) f ( x) F ( s) f (n) 2s' ' 2s' ' x 1 Case I k-space: Image Space: kz DFT ky Bernstein MA, Fain SB, and Riederer SJ, JMRI 14: 270-280 (2001) Methods: Sampling Case I: Zero-filled k-space: Image Space: kz FT ky Case II k-space: Image Space: kz FT ky Case II k-space: Image Space: kz FT ky Methods: Sampling Case III k-space: Image Space: kz FT ky Methods: Sampling Case III: Zero-Filled k-space: Image Space: kz FT ky