Indian Institute of Technology
Kharagpur
Department of Electrical Engineering |
Signal Processing and Machine
Learning
UNDERSTANDING K-SPACE IN MRI
IMAGE RECONSTRUCTION
Name : Arijit Sabui
Department : Signal Processing And Machine Learning
Year : 2nd Year ( Masters of Technology)
Roll No : 24EE65R09
Indian Institute of Technology
Kharagpur
Department of Electrical Engineering |
Signal Processing and Machine
Learning
Sr Number Content Page Number
1 Introduction to MRI and K-space 3
2 What is K-space? 4
3 Frequency and Phase Encoding 5
TABLE OF 4
5
K-space Sampling and Nyquist Criterion
Fourier and Inverse Fourier Transform
6
CONTENTS : 6 MRI Image Reconstruction 8
7 Under sampling and Aliasing 9
8 Conclusion 10
9 Thank You 11
Indian Institute of Technology
Kharagpur
Department of Electrical Engineering |
Signal Processing and Machine
INTRODUCTION TO MRI AND K-SPACE Learning
Magnetic Resonance Imaging (MRI):
A non-invasive imaging technique that provides excellent soft tissue contrast by
aligning hydrogen protons with a strong magnetic field and detecting their responses
to radiofrequency (RF) pulses.
Raw Data Collection:
The MRI scanner does not directly create images; instead, it captures radiofrequency
(RF) signals—known as echoes—that contain frequency-encoded spatial
information used to reconstruct detailed images.
K-space:
K-space is a mathematical matrix that stores raw MRI signal data, where each point
corresponds to a spatial frequency component, and the final image is generated by
applying a 2D Inverse Fourier Transform to this data.
Why K-space Matters:
Understanding k-space helps explain how MRI images are formed, why encoding is
required, and how scan time and image resolution are managed.
Indian Institute of Technology
Kharagpur
Department of Electrical Engineering |
Signal Processing and Machine
K SPACE Learning
What is K Space:
K-space is a data matrix in MRI where raw spatial frequency information is
stored during the scan.
It is not the image itself but contains the frequency domain representation of
the image.
Structure:
A 2D or 3D grid with:
Rows = Phase encoding
Columns = Frequency encoding
The centre of k-space stores low spatial frequencies → image contrast & gross
structure.
The edges of k-space store high spatial frequencies → image detail &
sharpness.
Key Points:
K-space is not spatial space — it’s a mathematical abstraction.
The final image is produced by applying a 2D Inverse Fourier Transform to k-
space data.
Indian Institute of Technology
Kharagpur
Department of Electrical Engineering |
Signal Processing and Machine
Why Encoding is Needed:
Learning
MRI uses magnetic field gradients to determine where in the body the
FREQUENCY MR signal comes from.
Two types of encoding are used to fill the k-space matrix: Phase
Encoding and Frequency Encoding.
AND PHASE Phase Encoding (Y-axis / Rows of k-space):
A gradient is applied briefly before signal acquisition.
ENCODING It changes the phase of spins along one direction (typically vertical).
For each line in k-space, a different phase gradient is applied.
Controls spatial resolution along the vertical axis of the image.
Frequency Encoding (X-axis / Columns of k-space):
Summary:
A gradient is applied during signal readout.
Encoding Type
Gradient Affects Axis Timing Causes spins in different locations (typically horizontal) to resonate at
Direction
different frequencies.
Frequency X (Readout) Horizontal During signal
Encoding readout The signal is Fourier-transformed along this axis to extract spatial
Phase Encoding Y (Pre-readout) Vertical Before signal info.
readout
Controls resolution along the horizontal axis.
Indian Institute of Technology
Kharagpur
Department of Electrical Engineering |
Signal Processing and Machine
Learning
K-space Sampling:
MRI fills k-space line-by-line, with each phase encoding step adding
K-SPACE one row of data.
Each row holds frequency-encoded signals; denser
SAMPLING AND sampling improves resolution but increases scan time.
Nyquist Criterion:
NYQUIST To avoid aliasing, the sampling rate must be at least twice the highest frequency in the
signal.
CRITERION In MRI: (spatial resolution)
FOV
Under sampling Consequence:
Under sampling violates Nyquist, leading to aliasing artifacts:
Image folds over itself
Detail is lost or misrepresented
Common in fast scans or when reducing scan time
Indian Institute of Technology
Kharagpur
Department of Electrical Engineering |
Signal Processing and Machine
Learning
Fourier Transform (FT):
Converts MRI signal from time/spatial domain to frequency
domain.
Used to organize raw signal data into k-space during
acquisition.
FOURIER AND Inverse Fourier Transform (IFT):
Converts data from k-space (frequency domain) back to
INVERSE FOURIER the image domain.
Produces the final spatial MRI image we see.
TRANSFORM Key Concept:
Each point in k-space contributes to the entire image.
Central k-space → general structure; outer k-space → fine
details.
∞ ∞
𝐼 ( 𝑥 , 𝑦 )= ∫ ∫ 𝑆(𝑘 𝑥 , 𝑘 𝑦 ) ∙ 𝑒 2 𝜋 𝑖 (𝑘 𝑥+𝑘 𝑦)
𝑥 𝑦
ⅆ 𝑘𝑥 ⅆ 𝑘𝑦
−∞ −∞
Indian Institute of Technology
Kharagpur
Department of Electrical Engineering |
Signal Processing and Machine
MRI IMAGE RECONSTRUCTION Learning
MRI Image Reconstruction :
Signal Acquisition:
MRI scanner collects raw signals (RF echoes) in response to gradient-encoded RF pulses.
K-space Filling:
Data from each phase and frequency encoding step is placed in the correct line/column of
RF SIGNALS DATA PLACED IN
k-space.
ACQUIRED K-SPACE
Inverse Fourier Transform:
A 2D IFT is applied to the fully or partially filled k-space to convert frequency data to
spatial image.
Final Image Formation:
The result is a grayscale image where brightness = signal intensity from hydrogen nuclei.
APPLY INVERSE
Quality Depends On: RECONSTRUCTED
FOURIER
MRI IMAGE
Image quality depends on: dense k-space sampling, high SNR, and compliance with TRANSFORM
Nyquist criterion.
Artifacts can arise from: motion, under sampling, or poor signal acquisition.
Indian Institute of Technology
Kharagpur
Department of Electrical Engineering |
Signal Processing and Machine
Under sampling: Learning
Happens when k-space is not fully sampled, especially
along the phase-encoding direction.
Used to reduce scan time, but comes at the cost of image
fidelity.
Aliasing: UNDER
A direct consequence of under sampling that violates the
Nyquist Criterion.
SAMPLING AND
High spatial frequencies are misinterpreted as lower ones
→ image overlaps or “wraps around”.
ALIASING
Impact on Image:
Appears as folded anatomy, ghosting, or streaks.
Common in fast imaging sequences or when Field of View
(FOV) is too small.
Relation to Nyquist: 𝑓 𝑠 <2 𝑓 𝑚𝑎𝑥 ⇒ 𝐴𝑙𝑖𝑎𝑠𝑖𝑛𝑔
Indian Institute of Technology
Kharagpur
Department of Electrical Engineering |
Signal Processing and Machine
Learning
CONCLUSION K-space is the foundational domain for MRI signal acquisition
and image formation.
Proper use of phase and frequency encoding, combined
with Nyquist compliant sampling, ensures accurate image
reconstruction.
Fourier and Inverse Fourier Transforms are essential for
transitioning between k-space and image space.
Under sampling saves time but risks introducing aliasing
artifacts if not handled correctly.
Understanding these principles allows for better image
quality, artifact management, and efficient scan design in clinical
MRI.
Indian Institute of Technology
Kharagpur
Department of Electrical Engineering |
Signal Processing and Machine
Learning
THANK YOU!
