Protocol Optimization and Image Quality in Magnetic Resonance Imaging

Fundamental Pillars of MR Image Quality

• Core Image Parameters: Magnetic resonance imaging protocols balance four interdependent parameters:

  • Signal-to-Noise Ratio ($\text{SNR}$): Ratio comparing true MR signal amplitude to random background noise.

  • Contrast-to-Noise Ratio ($\text{CNR}$): The difference in $\text{SNR}$ between adjacent tissues, allowing clear differentiation.

  • Spatial Resolution: The capacity to distinguish two small, closely spaced anatomical structures.

  • Scan Time: Total time required to acquire all necessary data lines in k-space.

• The Trade-Off Rule: Altering any single parameter directly impacts at least one other parameter. Improving spatial resolution through smaller voxels decreases $\text{SNR}$, while restoring $\text{SNR}$ typically increases scan time.

Signal-to-Noise Ratio ($\text{SNR}$)

• Signal vs. Noise:

  • Signal: Predictable voltage induced in the RF receiver coil by coherent transverse magnetization. Governed by scanner hardware, tissue properties, and sequence timing.

  • Noise: Unpredictable electrical fluctuations from thermal motion of charged particles in the patient (Brownian motion) and receiver electronics. Distributed across all frequencies.

• Hardware and Sample Factors:

  • Static Magnetic Field ($B<em>0$): Direct relationship. Higher $B</em>0$ increases proton alignment parallel to the field, enlarging the Net Magnetization Vector ($\text{NMV}$) and increasing $\text{SNR}$.

  • Trade-off: High field strengths deposit more radiofrequency energy into tissue, increasing Specific Absorption Rate ($\text{SAR}$) and heating risks.

  • Proton Density ($PD$): Direct relationship. Tissues with high mobile hydrogen content (such as CSF and fat) produce high signal, whereas low-proton tissues (such as cortical bone and lung) produce low signal.

  • Voxel Volume: Direct relationship. Voxel volume equals pixel area multiplied by slice thickness. Larger voxels enclose more protons, yielding higher $\text{SNR}$.

  • Trade-off: Larger voxels degrade spatial resolution and cause partial volume averaging.

  • Coil Selection: Use the smallest receiver coil that encompasses the region of interest ($\text{ROI}$) and place surface coils perpendicular to $B_0$. Smaller coils pick up less peripheral body noise, maximizing $\text{SNR}$.

  • Pulse Sequence Type: Spin Echo ($\text{SE}$) yields higher $\text{SNR}$ than Gradient Echo ($\text{GE}$) because the $180^\circ$ refocusing pulse corrects magnetic field inhomogeneities and prevents $T_2^*$ dephasing.

Timing, Averaging, and Bandwidth

• Repetition Time ($TR$): Direct relationship.

  • Longer $TR$ permits greater longitudinal recovery ($M<em>z$) via $T</em>1$ relaxation, leaving more magnetization available for subsequent RF excitation.

  • Trade-off: Longer $TR$ directly extends total scan time.

• Echo Time ($TE$): Inverse relationship.

  • Shorter $TE$ limits irreversible transverse decay ($T_2$ dephasing), preserving signal amplitude.

  • Trade-off: Extremely short $TE$ suppresses $T_2$ contrast weighting.

• Flip Angle ($\theta$): Direct relationship.

  • Larger flip angles (up to $90^\circ$ in spin echo) flip greater magnetization into the transverse plane to generate signal.

  • Trade-off: Low flip angles yield lower transverse signal, requiring compensatory increases in scan time or voxel dimensions.

• Number of Signal Averages ($NSA$ / $NEX$): Non-proportional direct relationship.

  • Signal sums coherently ($\propto NSA$) while random noise sums incoherently ($\propto \sqrt{NSA}$):
    $\text{SNR Factor} = \sqrt{NSA}$

  • Doubling $NSA$ increases $\text{SNR}$ by $40\%$ ($\sqrt{2} \approx 1.414$):
    $\text{New SNR} = \text{Baseline SNR} \times \sqrt{2}$

  • Quadrupling $NSA$ ($4 \times$) is required to double $\text{SNR}$ ($2 \times$).

  • Trade-off: Increasing $NSA$ lengthens scan time linearly.

• Receive Bandwidth ($rBW$): Inverse square-root relationship.

  • Narrowing the receiver filter excludes random background frequencies:
    $\text{SNR} \propto \frac{1}{\sqrt{rBW}}$

  • Reducing $rBW$ by $50\%$ increases $\text{SNR}$ by $40\%$.

  • Trade-offs: Increases minimum echo time ($TE_{\min}$), lengthens readout time, and increases chemical shift artifact.

Contrast-to-Noise Ratio ($\text{CNR}$)

• Definition: The numerical difference in $\text{SNR}$ between two adjacent tissues:
  $\text{CNR}<em>{AB} = \lvert \text{SNR}</em>A - \text{SNR}_B \rvert$

• Clinical Significance: High $\text{SNR}$ does not guarantee lesion conspicuity. If healthy tissue and diseased tissue share identical signal values, the pathology remains undetectable. $\text{CNR}$ enables tissue differentiation.

• Methods to Increase $\text{CNR}$:

  • Image Weighting: Select weighting where pathology differs from surrounding structures (e.g., $T_2$ weighting to make pathology bright against dark tissue).

  • Contrast Agents: Administer intravenous contrast media (such as gadolinium) to selectively alter tissue relaxation rates and highlight pathology.

  • Chemical Pre-saturation: Apply RF saturation pulses to null signal from unwanted normal tissues (such as fat saturation) so pathology stands out.

  • Magnetization Transfer Contrast ($\text{MTC}$): Apply off-resonance RF pulses to saturate bound protons, suppressing normal background tissue and boosting contrast with free-water pathology.

Spatial Resolution and Scan Time

• Spatial Resolution:

  • Determined by voxel dimensions: slice thickness, Field of View ($\text{FOV}$), and matrix dimensions.

  • Increasing matrix dimensions (e.g., from $128 \times 128$ to $256 \times 256$) creates smaller pixels, improving spatial resolution at the expense of $\text{SNR}$.

• Scan Time Calculations:

  • Conventional Spin Echo ($\text{CSE}$):
    $\text{Scan Time} = TR \times \text{Phase Encodings} \times NSA$

  • Fast Spin Echo ($\text{FSE}$):
    $\text{Scan Time} = \frac{TR \times \text{Phase Encodings} \times NSA}{\text{Echo Train Length}}$

• Methods to Shorten Scan Time:

  • Use the shortest practical $TR$.

  • Use a coarser phase-encoding matrix.

  • Reduce $NSA$.

  • Use a rectangular $\text{FOV}$ to decrease phase-encoding steps.