Protocol Design & Sequence Selection

A protocol is an argument

Every MRI examination is a structured argument built to answer one clinical question. The referrer asks something specific, demyelination versus not, schwannoma versus not, meniscal tear versus not, and the protocol is the chain of evidence you assemble in response. Each sequence is one premise: it isolates a tissue property, in a plane, with a contrast mechanism, so that the abnormality you are hunting becomes conspicuous against everything else. A well-designed protocol contains no redundant premises and no missing ones. It fits inside a finite scan-time budget (patient tolerance, throughput, motion) and a finite SAR budget (regulatory and thermal limits), and it spends those budgets where the diagnostic yield is highest.

This lesson treats protocol design as engineering rather than recipe-following. We will enumerate the parameter levers and the trade-offs each one forces, derive the time and signal-to-noise relationships that govern those trade-offs, build a rational framework for choosing among fat-suppression strategies, and then work three indications from clinical question to sequence list. Throughout, the goal is judgment you can defend, not memorized tables, because real scanners, patients, and questions never quite match the template.

The parameter levers and their trade-offs

Protocol parameters fall into two groups. Contrast levers, the plane, weighting (T1, T2, PD, T2*, diffusion), fat suppression, and gadolinium administration, decide what becomes bright or dark. Quality-versus-time levers, slice thickness, field of view (FOV), acquisition matrix, receiver bandwidth, and number of signal averages (NSA), decide how clean and how fine the picture is and how long it takes. The contrast levers are chosen from the clinical question; the quality levers are then tuned to make the answer readable within the budget.

The signal-to-noise ratio (SNR) of a voxel is the unifying currency. To first order it is proportional to the voxel volume and to the square root of the total sampling time, and inversely proportional to the square root of the receiver bandwidth per pixel.

Two consequences follow immediately. First, resolution is expensive in SNR: halving each in-plane dimension to double resolution cuts voxel volume to one quarter, so SNR drops fourfold unless you spend time to recover it. Second, the time cost is set mostly by the phase-encode count and the averages. For a conventional 2D acquisition the scan time is

Reading these together gives the levers you actually pull when a protocol overruns. Doubling NSA buys $\sqrt{2}\approx 1.41$ times the SNR for twice the time, a poor exchange used only to fight motion. Halving the bandwidth also buys $\sqrt{2}$ SNR at no time cost, but doubles the echo spacing, worsens chemical-shift misregistration, and can lengthen the minimum TE. Raising the ETL shortens the scan proportionally but blurs the image because later echoes are T2-weighted and fill the high-frequency periphery of k-space. Partial Fourier and parallel imaging (acceleration $R$) cut $N_{PE}$ directly; parallel imaging costs SNR by roughly $1/(g\sqrt{R})$, where the g-factor $g \geq 1$ reflects coil geometry.

| Lever | Effect on SNR | Effect on time | Main penalty | | --- | --- | --- | --- | | Smaller voxel (finer matrix or thinner slice) | Lower (voxel volume) | Higher (more N_PE) | Noise, longer scan | | More averages (NSA) | Higher as square root | Linearly higher | Time; only partial motion benefit | | Lower receiver bandwidth | Higher as square root | Slightly higher TE | Chemical shift, longer echo spacing | | Longer echo train (ETL) | Roughly neutral | Lower | Blurring, T2 contrast drift | | Parallel imaging (R) | Lower by g times root R | Lower | Reconstruction artifact, g-factor noise |

Choosing a fat-suppression strategy

Fat suppression is one of the most consequential and most misused levers. Fat is bright on both T1- and T2-weighted fast spin-echo images, and it can mimic, mask, or be mistaken for pathology. Four physically distinct strategies exist, and they fail in different ways, so the right choice depends on field homogeneity, whether contrast has been given, and how far off isocenter the anatomy lies.

Chemical-shift selective saturation (CHESS / fat-sat)

Fat and water protons resonate about 3.5 ppm apart, which at 1.5 T is roughly 220 Hz and at 3 T roughly 440 Hz. A frequency-selective $90^{\circ}$ pulse tuned to the fat peak, followed by a spoiler gradient, nulls fat longitudinal magnetization before the imaging excitation. It is spectrally specific and works well post-gadolinium because it suppresses fat without touching the T1 shortening of enhancing tissue. Its weakness is dependence on a uniform $B_0$: a few ppm of off-resonance, common near the neck, shoulders, or any large FOV, shifts the saturation band onto water and produces patchy, asymmetric failure.

Short-tau inversion recovery (STIR)

STIR exploits T1 rather than chemistry. An inversion pulse is followed by an inversion time TI chosen so that the recovering fat magnetization crosses zero at the moment of excitation. The null time is

At 1.5 T this gives $TI \approx 150$–$170$ ms. Because nulling depends on T1, not resonant frequency, STIR is gloriously insensitive to $B_0$ inhomogeneity, which is why it is the workhorse for large or off-isocenter fields such as the whole spine, the brachial plexus, or extremities far from magnet center. Two cautions follow. First, the null is non-selective for tissues sharing fat's T1, so STIR also suppresses signal from anything with a similar T1, including enhancing tissue after gadolinium, which is why STIR is contraindicated post-contrast. Second, the inversion recovery costs SNR and lengthens TR.

Dixon and water excitation

Dixon methods acquire in-phase and opposed-phase echoes and solve algebraically for separate water-only and fat-only images, yielding four contrasts (in-phase, opposed-phase, water, fat) from one acquisition. Because the separation is computed rather than spectrally selected, two-point and three-point Dixon are robust to moderate $B_0$ inhomogeneity, give uniform fat suppression over large fields, and behave well post-contrast, at the cost of slightly longer acquisition and dependence on a stable phase model. Water excitation uses a composite binomial pulse (for example 1-1 or 1-2-1) that delivers flip angle to water while returning fat to the longitudinal axis, avoiding a separate saturation pulse; it is fast and SAR-light and is favored in 3D cartilage imaging, but it too is field-dependent.

| Strategy | Mechanism | B0 sensitivity | Post-contrast? | Best use | | --- | --- | --- | --- | --- | | CHESS fat-sat | Frequency-selective saturation | High (fails off-resonance) | Yes, preferred | Homogeneous fields, post-Gd at isocenter | | STIR | T1 null via inversion time | Low (very robust) | No (nulls enhancement) | Large FOV, spine, off-isocenter, metal-adjacent | | Dixon | In/opposed-phase decomposition | Low to moderate | Yes | Large FOV uniform fat-sat, single-acquisition multi-contrast | | Water excitation | Binomial selective excitation | Moderate | Yes | 3D cartilage, SAR-limited 3 T |

Budgeting time, SAR, and contrast

A protocol must close two budgets. The time budget is the sum of Eq. 17.2 over all sequences plus localizers, calibration, and contrast-injection delays, and it competes against patient tolerance (often 30–45 minutes before motion degrades everything) and department throughput. The SAR budget is bounded by IEC limits, with whole-body SAR capped at 4 W/kg in first-level controlled mode and 2 W/kg in normal mode. SAR rises steeply with the radiofrequency burden of a sequence.

The quadratic dependence on $B_0$ is why a sequence that is comfortable at 1.5 T can hit the SAR ceiling at 3 T, forcing longer TR, lower refocusing flip angles (hyperecho or variable-flip 3D FSE such as SPACE/CUBE/VISTA), or fewer slices per TR. SAR-heavy ingredients are long echo trains, $180^{\circ}$ refocusing, magnetization-transfer pulses, and short TR. The clinical decision to give gadolinium is its own budget item: it adds an injection, a delay, and post-contrast sequences, and is justified when enhancement materially changes the answer, blood-brain-barrier breakdown in active demyelination or tumor, leptomeningeal or epidural disease, infection, and most soft-tissue or marrow masses. It is usually unnecessary for a routine first-presentation lumbar disc or a simple meniscal evaluation.

Justifying weighting choices interactively

Before committing to a sequence list, it helps to see how TR and TE move tissue contrast, because the weighting you choose is the lever that makes the target lesion conspicuous. Short TR and short TE give T1 weighting (fat and subacute hemorrhage bright, fluid dark); long TR and long TE give T2 weighting (fluid and most pathology bright); long TR and short TE give proton-density weighting that maximizes anatomic signal. Use the playground below to drive a lesion against a background and confirm the weighting that maximizes contrast for the question you have in mind.

Notice the practical lessons the playground makes concrete. Edema, gliosis, and most tumors lengthen both T1 and T2, so they are dark on T1 and bright on T2/FLAIR; this is why a fluid-sensitive sequence is the cornerstone of nearly every protocol. Watch how at very long TE the overall signal collapses (T2 decay) so contrast-to-noise, not raw contrast, is what you optimize. And note how adding a T1-shortening contrast agent flips an enhancing lesion from dark to bright on T1, the basis for every post-gadolinium series.

Worked examples: indication to sequence list

MS / demyelination brain

The question is dissemination of demyelinating lesions in space and time, including activity. The argument: a 3D or 2D FLAIR to make periventricular, juxtacortical, and infratentorial plaques conspicuous against suppressed CSF; a T2 FSE for confirmation and posterior-fossa detail; sagittal FLAIR to show the perpendicular callosal-radial Dawson fingers; DWI to flag acute lesions and exclude mimics; and pre- and post-gadolinium T1 (often 3D) where enhancement signals an active, barrier-disrupted lesion. Fat suppression is rarely needed intracranially, so the fat-sat debate is mostly moot here; the budget is spent instead on thin-section 3D FLAIR for lesion counting.

Internal auditory canal for vestibular schwannoma

The question is a small enhancing mass within the IAC or cerebellopontine angle. The argument hinges on resolution and CNR: a high-resolution heavily T2-weighted 3D sequence (CISS/FIESTA-C) renders the cisternal nerves as dark filling defects in bright CSF at sub-millimeter voxels, and thin post-gadolinium 3D T1 detects the enhancing tumor. Because the IAC sits well off isocenter near bone-air interfaces, frequency-selective fat-sat is unreliable; this is exactly where Dixon or a non-fat-suppressed high-resolution acquisition earns its place. The protocol deliberately sacrifices coverage for resolution, the opposite of the MS protocol's priorities.

Routine knee or lumbar spine

For the knee the question is internal derangement: sagittal and coronal PD FSE (short TE) to show meniscus and ligament morphology, paired with fat-suppressed T2 or PD (or STIR) to reveal marrow edema, fluid, and cartilage; axial fat-suppressed images for patellofemoral cartilage and retinacula. Gadolinium is reserved for tumor, infection, or postoperative residual-tear questions (MR arthrography). For the lumbar spine the question is usually radicular compression: sagittal T1 and T2, axial T2 through the discs, and a sagittal STIR to catch marrow edema, fracture, or infiltration. STIR, not fat-sat, is chosen because the large head-to-foot FOV defeats frequency-selective suppression. Contrast is added only for suspected infection, neoplasm, or the postoperative back to separate scar from recurrent disc.

Adapting the protocol to the patient

The template bends to reality. For implants and metal, switch to STIR for fat suppression, raise the receiver bandwidth to shrink chemical-shift and susceptibility displacement, increase the ETL, use thinner slices, and add view-angle-tilting or dedicated metal-artifact-reduction sequences (MAVRIC/SEMAC); avoid gradient-echo, which is exquisitely susceptibility-sensitive. For motion-prone or uncooperative patients, shorten the exam ruthlessly to the essential sequences, raise the ETL and use parallel imaging to finish faster, deploy radial or PROPELLER/BLADE acquisitions whose oversampled center averages out motion, and use single-shot fast spin-echo for breath-hold-free coverage. For claustrophobia, use a wide-bore or open scanner, prone positioning, feet-first entry, mirrors, and the shortest viable protocol, and recognize that a completed limited study beats an aborted complete one.