Master the liver MRI protocol: LAVA/THRIVE/VIBE sequencing, gadoxetate dynamics, navigator-gated motion control, LI-RADS reporting, and the pitfalls that separate a diagnostic study from a repeat scan.

Liver MRI Protocol: 10 Critical Steps for Dynamic Multiphasic Imaging

Introduction

The liver MRI protocol built around dynamic multiphasic contrast-enhanced imaging is one of the most technically demanding and clinically consequential studies performed in an abdominal MRI suite. Unlike a static anatomical sequence, this protocol asks the scanner, the injector, and the patient’s own diaphragm to move in near-perfect synchrony across a compressed window of only 15–20 seconds per breath-hold, repeated three or four times, each capturing a distinct and irreversible moment of contrast kinetics through the dual hepatic arterial and portal venous blood supply.

Get the timing wrong by even a few seconds, or let the patient’s breathing drift by a few millimetres, and an entire phase can be rendered non-diagnostic — with no way to reacquire it once the contrast bolus has passed. This is why the liver protocol sits at the intersection of physics, physiology, and workflow discipline in a way that few other MRI examinations do.

This article walks through every stage of the protocol: the anatomy that dictates coil and slab positioning, the relaxation properties that govern sequence weighting, the ten-step scanning workflow, the contrast injection dynamics, radiation-free but power-limited RF safety considerations, the LI-RADS reporting framework that standardises HCC risk stratification, the ten pathologies most frequently characterised on this protocol, and — critically — the pitfalls that trip up radiographers, radiologists, and referring physicians at each stage of the diagnostic chain.

The stakes of getting this examination right extend well beyond a single radiology report. In cirrhotic patients under active HCC surveillance, this protocol frequently determines whether a patient proceeds directly to curative-intent treatment, requires a confirmatory biopsy, or is referred back into a longer surveillance interval — decisions with direct, measurable impact on survival. A technically compromised examination does not merely produce a suboptimal image; it can silently propagate uncertainty all the way to the multidisciplinary tumour board, where clinicians who were not present at the scanner must make treatment decisions based on the images they are given.

Anatomy of the liver

The liver is the largest solid abdominal viscus, occupying the right upper quadrant beneath the diaphragm and extending across the midline into the left upper quadrant. It is conventionally divided using the Couinaud classification into eight functionally independent segments, each defined by its own portal pedicle (portal vein, hepatic artery, and bile duct branch) and drained by one of three hepatic veins into the inferior vena cava (IVC). This segmental anatomy is not academic trivia — it is the language surgeons use to plan resections, and the radiologist’s report must localise every lesion to a specific segment or segments.

Segments II and III form the left lateral section, segment IV (further divided into IVa and IVb) sits medially adjacent to the gallbladder fossa, and segments V through VIII comprise the right lobe. The caudate lobe (segment I) is anatomically and functionally distinct, receiving blood supply from both the right and left portal pedicles and draining directly into the IVC via short hepatic veins — a feature that makes it relatively spared in some forms of Budd-Chiari syndrome and relevant when staging tumours abutting the IVC.

Dual blood supply and the basis of multiphasic imaging

The liver is unique among abdominal organs in receiving a dual blood supply: approximately 25–30% from the hepatic artery (oxygen-rich) and 70–75% from the portal vein (nutrient-rich, draining the gut and spleen). This dual supply is the entire physiological basis for the multiphasic protocol. Hypervascular lesions such as HCC derive their blood supply almost exclusively from the hepatic artery, while normal liver parenchyma is predominantly portally perfused. The arterial phase therefore exploits a 5–8 second physiological window in which a tumour “lights up” against a background liver that has not yet been perfused by the slower portal circulation.

Biliary tree and vascular landmarks

The intrahepatic bile ducts converge at the porta hepatis to form the common hepatic duct, joined by the cystic duct to form the common bile duct. The main portal vein bifurcates at the porta hepatis into right and left branches, while the proper hepatic artery — arising from the celiac trunk via the common hepatic artery — divides similarly, though anatomical variants (replaced right hepatic artery from the superior mesenteric artery, replaced left hepatic artery from the left gastric artery) are present in up to 20–45% of patients and must be recognised, particularly for surgical and transplant planning.

Relationship to the diaphragm and respiratory motion

The liver’s superior surface is in direct contact with the diaphragm, moving craniocaudally by 1–4 cm during quiet respiration and considerably more during deep or irregular breathing. This intimate diaphragmatic relationship is the anatomical root cause of the protocol’s signature artefact: because the liver moves substantially with every breath, any inconsistency in breath-hold depth between phases — or any involuntary breath during a “held” acquisition — directly translates into misregistration, blurring, and ghosting on the images.

Lymphatic drainage, innervation, and adjacent organ relationships

Hepatic lymphatic drainage follows two principal pathways: superficial subcapsular lymphatics draining toward the diaphragmatic and mediastinal nodal chains, and deeper parenchymal lymphatics accompanying the portal triads toward the hepatoduodenal ligament and porta hepatis nodes. This dual drainage pattern is clinically relevant when staging hepatobiliary malignancy, since nodal spread can bypass the porta hepatis entirely in tumours arising near the bare area of the liver, presenting instead with mediastinal or diaphragmatic nodal involvement that falls outside a standard abdominal MRI field of view.

The liver’s visceral peritoneal relationships also matter directly to protocol planning: the bare area, where the liver is in direct contact with the diaphragm without an intervening peritoneal layer, is a common site of diaphragmatic crus artefact and susceptibility-related signal dropout on gradient-echo sequences, and radiographers should anticipate reduced image quality here regardless of breath-hold fidelity. Anteriorly, the liver’s relationship to the anterior abdominal wall and costal margin dictates optimal coil placement, while posteriorly its proximity to the right kidney, right adrenal gland, and inferior vena cava means that adequate coverage inferiorly is essential to avoid missing caudate lobe or segment VI/VII lesions that abut these structures.

Autonomic innervation of the liver arises from the hepatic plexus, itself derived from the celiac plexus, and travels alongside the hepatic artery into the parenchyma — a detail with limited direct imaging relevance but important context for understanding referred right upper quadrant and right shoulder pain patterns seen clinically in capsular distension or acute hepatitis, which frequently prompts the initial imaging referral that leads to this protocol being performed.

This arterial anatomy is not purely diagnostic in relevance. In patients ultimately categorised as LR-5 (definite HCC) who are not surgical or transplant candidates, transarterial chemoembolisation delivers a chemotoxic drug-in-oil emulsion selectively into the tumour-feeding hepatic artery branch, exploiting the same dual-blood-supply physiology that makes the arterial phase of this MRI protocol diagnostic in the first place. Consistent, well-mixed emulsion preparation is essential to procedural success, since an unstable emulsion can separate prematurely within the catheter, reducing tumoricidal drug delivery and increasing non-target embolisation risk.

MR tissue relaxation values

Understanding native T1 and T2 relaxation times of hepatic and adjacent structures at both field strengths is essential for protocol optimisation, sequence weighting, and troubleshooting unexpected signal behaviour. Values below are representative ranges compiled from field-strength-specific literature and vendor phantom studies; individual scanner calibration and patient factors (iron overload, fibrosis, steatosis) will shift these figures.

Field-strength-dependent T1 prolongation at 3.0T relative to 1.5T is a consistent and predictable phenomenon across all the tissues listed above, driven by the physics of proton relaxation at higher static magnetic field. This has a direct practical consequence: pulse sequence parameters optimised at 1.5T cannot simply be copied to a 3.0T system without adjustment, since the underlying T1 contrast-to-noise relationship between tumour and background liver shifts with field strength, which is why the scanner comparison table below specifies distinct flip angle and timing recommendations for each field strength rather than a single universal protocol.

Scanning technique

The dynamic multiphasic liver MRI protocol is executed as a tightly sequenced ten-step workflow. Every step feeds directly into the next, and a shortcut taken at step 3 (coil placement) or step 5 (localiser/navigator planning) will surface as a diagnostic failure at step 8 (arterial phase acquisition), by which point the contrast bolus cannot be recovered.

Total examination time for a complete protocol, including pre-contrast sequences, dynamic phases, and an optional hepatobiliary phase, typically ranges from 30 to 45 minutes on the table, considerably longer than many other single-region MRI examinations. This extended duration compounds the cumulative fatigue effect on breath-hold reliability discussed further below and makes efficient, well-rehearsed workflow execution a genuine determinant of diagnostic success rather than simply a matter of departmental throughput.

1. Patient preparation and breath-hold coaching. Explain the sequence of breath-holds in advance — most protocols require 4–6 discrete breath-holds of 12–20 seconds each. Rehearse the exact instruction phrasing (“breathe in, breathe out, hold”) so the patient reproduces an identical lung volume at each phase, not merely “any” held breath.
2. Screening and IV access. Confirm renal function status (eGFR) where a linear GBCA is being considered, secure an 18–20G cannula in the antecubital fossa (right arm preferred to avoid left brachiocephalic vein compression artefact), and connect a power injector rated for the planned flow rate.
3. Coil selection and positioning. A torso phased-array coil with sufficient posterior and anterior element density is positioned to cover the dome of the diaphragm superiorly through the lower pole of the right kidney inferiorly, centred over the xiphisternum.
4. Localiser and coverage planning. Triplanar localisers define the imaging slab; the field of view must include the entire liver from diaphragmatic dome to inferior tip, with adequate margin for potential accessory or ectopic hepatic tissue.
5. Navigator/respiratory reference placement. For navigator-gated acquisitions, the tracking pencil-beam or BioMatrix-type sensor is positioned at the dome of the right hemidiaphragm, avoiding lung parenchyma and large vessels that would corrupt the respiratory waveform.
6. Pre-contrast sequences. Acquire axial and coronal T2-weighted single-shot fast spin echo, in-phase/opposed-phase T1 GRE (for fat and iron quantification), and diffusion-weighted imaging (b=50, 400, 800 s/mm²) — all before contrast, as DWI is contrast-sensitive and must precede injection.
7. Pre-contrast fat-suppressed 3D T1 GRE mask. A non-contrast 3D LAVA/THRIVE/VIBE acquisition establishes the subtraction mask and confirms breath-hold reproducibility before the irrecoverable contrast bolus is given.
8. Bolus timing determination. Use either a timing bolus (small test injection with low-dose dynamic monitoring) or automated bolus-triggering (fluoroscopic trigger on aortic enhancement) to determine the precise scan-delay for the arterial phase.
9. Dynamic contrast-enhanced acquisition. Inject the full contrast dose and acquire the arterial phase at the predetermined delay, immediately followed by portal venous phase (60–70 seconds post-injection) and delayed/equilibrium phase (3 minutes), with an optional hepatobiliary phase at 20 minutes for gadoxetate disodium.
10. Post-processing and quality review. Generate subtraction images (post-minus-pre contrast), review each phase in real time for motion degradation, and repeat only the affected phase if a free-breathing or single-shot fallback sequence is available before the patient leaves the scanner.

Each of these ten steps carries a specific failure mode that a well-trained radiographer learns to anticipate. Step 1 failures (inadequate breath-hold coaching) manifest as progressively degraded image quality across successive phases as patient fatigue accumulates — the fourth or fifth breath-hold of a session is measurably less reliable than the first, which is precisely why the highest-priority arterial phase should never be scheduled as the final acquisition of a long protocol. Step 5 failures (poor navigator placement) produce an unstable or noisy respiratory trace that either triggers acquisition at the wrong point in the respiratory cycle or fails to trigger reliably at all, forcing a fallback to free-breathing or breath-hold-only technique mid-examination.

Step 8 (bolus timing) deserves particular emphasis because it is the step most frequently compressed under time pressure. A rushed or omitted timing bolus, replaced by a fixed population-average delay, will work adequately for a patient with normal cardiac output but will systematically mistime the arterial phase in patients with reduced ejection fraction, arrhythmia, or significant peripheral vascular disease — precisely the multimorbid oncology and cirrhotic populations most commonly referred for this examination. Individualised bolus timing is therefore not a refinement reserved for difficult cases; it is baseline good practice for this protocol.

Contrast media protocol

Contrast administration is the physiological engine of the dynamic liver MRI protocol. The standard regimen for an extracellular gadolinium-based contrast agent (GBCA) or gadoxetate disodium is 10–15 mL (0.1 mmol/kg) delivered intravenously at 3.0 mL/s, immediately followed by a 100 mL normal saline chaser also administered at 3.0 mL/s to fully clear the injection line and tubing dead-space, ensuring a tight, compact contrast bolus rather than a smeared, prolonged one.

Extracellular versus hepatobiliary agents

Gadolinium-based contrast agents used in this protocol fall into two functional classes. Extracellular agents (e.g., gadobutrol, gadoterate meglumine, gadobenate dimeglumine at extracellular dosing) distribute into the vascular and interstitial space and are excreted renally, providing robust arterial, portal venous, and delayed-phase kinetics without a hepatobiliary phase option. Hepatocyte-specific agents, principally gadoxetate disodium, combine conventional extracellular distribution in the early dynamic phases with subsequent active uptake by functioning hepatocytes via the OATP1B1/B3 transporter, enabling the additional 20-minute hepatobiliary phase that markedly improves sensitivity for well-differentiated HCC and characterisation of focal nodular hyperplasia, at the cost of a lower per-kilogram gadolinium concentration and a correspondingly higher risk of transient severe motion when injected rapidly and undiluted.

The choice between agent classes is typically an institutional protocol decision rather than a per-patient one, though renal function, prior reaction history, and the specific clinical question (e.g., FNH versus adenoma characterisation strongly favours a hepatobiliary agent) all inform selection. Whichever agent is chosen, the injection parameters described below — rate, volume, and saline chase — remain the primary levers available to the radiographer to control bolus geometry and arterial phase image quality.

The weight-based 0.1 mmol/kg dosing standard reflects decades of pharmacokinetic study balancing adequate T1-shortening effect against gadolinium exposure minimisation, consistent with the “as low as reasonably achievable” principle increasingly applied to gadolinium administration following heightened awareness of tissue gadolinium retention. The 3.0 mL/s injection rate represents a practical compromise for extracellular agents: fast enough to produce a compact, well-defined bolus for reliable arterial-phase timing, yet slow enough to remain comfortably below typical peripheral IV cannula flow tolerances and avoid extravasation risk in the antecubital veins most commonly used for this examination.

Three (or four, with gadoxetate) temporally distinct phases are acquired:

• Arterial phase (15–35 seconds post-injection, patient- and cardiac-output-dependent): captures peak hepatic arterial enhancement while portal venous enhancement remains minimal — the phase most sensitive to hypervascular lesions such as HCC and most vulnerable to transient severe motion (TSM).
• Portal venous phase (60–70 seconds): near-peak parenchymal and portal venous enhancement, optimal for detecting hypovascular metastases and assessing portal vein patency.
• Delayed/equilibrium phase (3 minutes): demonstrates washout kinetics of hypervascular lesions and capsular enhancement — a key LI-RADS ancillary feature.
• Hepatobiliary phase (20 minutes, gadoxetate disodium only): exploits OATP1 transporter uptake by functioning hepatocytes, providing unmatched sensitivity for well-differentiated HCC and focal nodular hyperplasia characterisation.

Scheduling the optional 20-minute hepatobiliary phase has real workflow implications that departments must plan for explicitly. Unlike the arterial, portal venous, and delayed phases, which occupy the patient continuously on the table, the 20-minute interval before the hepatobiliary phase is frequently used to scan a second patient on the same magnet, requiring the first patient to remain positioned (or be briefly removed and precisely repositioned) for the final acquisition. Departments that fail to plan this interval explicitly often default to skipping the hepatobiliary phase under time pressure, forfeiting its considerable diagnostic value for FNH and well-differentiated HCC characterisation described in the LI-RADS section of this article.

Gadoxetate disodium carries a well-documented risk of transient severe motion (TSM) — an involuntary, exaggerated respiratory response occurring seconds after rapid, undiluted injection, reported in 2.4–18% of examinations. A large cohort study of 1,413 patients found that diluting gadoxetate 1:1 with saline and slowing the injection rate to approximately 1 mL/s produced artefact-free arterial phase images in 77.8% of examinations, with only 5.4% showing moderate residual artefact — establishing dilution and slow-rate injection as an evidence-based mitigation strategy that requires a power injector capable of accurate, reproducible low-rate delivery.^[5]

Specific absorption rate

Specific absorption rate (SAR) quantifies RF energy deposition in tissue (W/kg) and becomes a genuine rate-limiting factor for the liver protocol at 3.0T, where multiple rapid breath-hold 3D GRE acquisitions are stacked back-to-back within a short overall examination time. Excessive SAR forces the scanner to either lengthen TR (undermining breath-hold feasibility) or reduce flip angle (reducing T1 contrast), both of which directly compromise diagnostic quality of the arterial phase.

SAR management in this protocol is complicated by the fact that the dynamic phases are, by clinical necessity, acquired in rapid succession with minimal inter-phase delay — the arterial-to-portal-venous transition window is deliberately kept short to maintain temporal resolution of the enhancement curve. This compresses RF duty cycle into a short overall window and can push cumulative SAR toward regulatory limits more quickly at 3.0T than in a typical single-sequence abdominal examination, particularly in larger patients where absolute power deposition (proportional to body mass) is higher for an equivalent per-kilogram SAR figure.

Five evidence-based strategies reduce SAR while preserving image quality, aligned to EC RP 185, AAPM, and ICRP guidance on medical RF exposure:

1. Increase parallel imaging acceleration (SENSE/GRAPPA/ARC), which reduces the number of RF pulses required per unit time without proportionally sacrificing spatial resolution.
2. Lower the flip angle modestly (e.g., 12° → 10° at 3.0T) where T1 contrast headroom allows, particularly for the pre-contrast mask acquisition.
3. Extend TR marginally between consecutive breath-hold phases to allow RF energy dissipation, balanced against total scan-time constraints.
4. Use hyperecho or low-SAR refocusing pulse trains for any accompanying fast spin-echo sequences (e.g., T2 SSFSE) acquired within the same session.
5. Employ vendor-specific SAR-optimised pulse design (e.g., variable-rate selective excitation, VERSE) which reshapes RF pulses to reduce peak power while preserving slice profile fidelity.

In practice, the most effective single intervention is usually the first: increasing parallel imaging acceleration simultaneously reduces both scan time (helping breath-hold feasibility) and SAR (helping regulatory compliance), making it the preferred first lever before resorting to flip angle reduction, which directly trades away T1 contrast — the very property the arterial phase depends on to detect hypervascular lesions.

LI-RADS: the Liver Imaging Reporting and Data System

No discussion of the dynamic liver MRI protocol is complete without a detailed treatment of the Liver Imaging Reporting and Data System (LI-RADS), the standardised lexicon and algorithm that governs how every arterial-phase, portal-venous-phase, and delayed-phase image acquired under this protocol is ultimately translated into a reproducible, actionable diagnostic category. LI-RADS is not a peripheral add-on to the liver protocol — it is the reason the protocol is built the way it is. Every technical decision described in the preceding sections, from navigator-gated breath-hold consistency to precise arterial-phase bolus timing, exists in service of generating images that meet LI-RADS technical adequacy criteria.

Origin, governance, and purpose

LI-RADS was developed and is maintained by the American College of Radiology (ACR), first published in 2011 and iteratively refined through several major versions, with LI-RADS v2018 remaining the most widely implemented framework in current clinical practice at the time of writing, alongside ongoing update cycles that incorporate emerging evidence on ancillary features and treatment response assessment. The system was created to solve a specific and costly clinical problem: prior to standardisation, radiologists describing focal liver lesions in at-risk patients used inconsistent, subjective terminology — “probably benign,” “suspicious,” “cannot exclude HCC” — that varied enormously between readers and institutions, leading to unpredictable downstream management, unnecessary biopsies, and delayed treatment of true hepatocellular carcinoma (HCC).

LI-RADS addresses this by defining a strict, reproducible lexicon of major and ancillary imaging features, combining them through an explicit algorithm, and assigning each observation to one of a small number of standardised categories that map directly onto probability of malignancy and, critically, onto a specific downstream management pathway. This transforms the radiology report from a narrative opinion into a structured, actionable data point that surgeons, hepatologists, and transplant teams can act on without needing to reinterpret free-text impressions.

Target population: who LI-RADS applies to

LI-RADS is deliberately scoped to a specific at-risk population and should not be applied indiscriminately to every liver lesion encountered in general practice. The system is validated for adult patients with:

• Cirrhosis of any aetiology (viral, alcohol-related, metabolic-associated steatotic liver disease, autoimmune, or other);
• Chronic hepatitis B infection, even in the absence of cirrhosis;
• Current or prior HCC, including patients under post-treatment surveillance.

Patients outside this risk profile — for example, a young patient with no liver disease risk factors and an incidentally discovered lesion — should be evaluated using standard, non-LI-RADS descriptive reporting, since the pre-test probability of HCC in that population is fundamentally different and the LI-RADS algorithm’s predictive performance has not been validated in that context.

Technical requirements for a LI-RADS-compliant MRI

To generate a valid LI-RADS category, the MRI examination must satisfy several technical prerequisites that map directly onto the ten-step scanning workflow described above:

• Multiphasic dynamic contrast-enhanced acquisition including, at minimum, late arterial, portal venous, and delayed phases — precisely the three-to-four phase protocol detailed in the contrast media section.
• Adequate late arterial phase timing, defined as the phase in which hepatic arteries are well opacified but hepatic veins show minimal-to-no enhancement — achieved through bolus-tracking or timing-bolus techniques rather than a fixed empirical delay.
• Complete hepatic coverage from diaphragmatic dome to inferior liver margin without missing segments due to slab positioning errors.
• Adequate spatial and temporal resolution to detect lesions ≥10 mm with confidence and to characterise washout kinetics between phases.
• Motion-free or motion-minimised acquisition, since even moderate respiratory misregistration between phases can obscure or simulate arterial phase hyperenhancement (APHE) and washout — the two pivotal major features on which the entire algorithm turns.

Major imaging features

LI-RADS categorisation is built primarily on four major features, each with a precise operational definition to minimise inter-reader variability:

These four features are then combined with lesion size using the LI-RADS diagnostic table: a lesion <20 mm with APHE plus one additional major feature (washout, capsule, or threshold growth) is categorised as LR-5 (definitely HCC), while a lesion ≥20 mm with APHE alone, without any other major feature, already qualifies as LR-5 given the higher pre-test probability associated with larger arterially hyperenhancing lesions in cirrhotic liver.

Ancillary features

Beyond the four major features, LI-RADS defines a set of ancillary features that can nudge a category up or down within defined limits but cannot, on their own, generate an LR-5 designation. These are subdivided into features favouring malignancy in general, features favouring HCC specifically, and features favouring benignity.

The LI-RADS categories

The full category system spans from definite benignity to definite malignancy, plus special categories for indeterminate observations, tumour in vein, and technically inadequate studies:

The clinical elegance of this system is that LR-5 carries a positive predictive value for HCC exceeding 95% in most validation cohorts, high enough that major transplant and oncology guidelines permit proceeding directly to treatment — including transplant listing in eligible candidates — without requiring a confirmatory biopsy, a paradigm shift that spares patients an invasive procedure with attendant seeding risk while accelerating time to treatment.

Hepatobiliary agents and the LI-RADS algorithm

When gadoxetate disodium is used rather than a purely extracellular agent, LI-RADS requires a specific technical modification: nonperipheral washout must be assessed only on the portal venous phase, not on the transitional or hepatobiliary phase, because normal liver parenchyma begins actively taking up gadoxetate via OATP1 transporters during the transitional phase, which can create an artefactual “pseudo-washout” appearance in lesions that lack true HCC-type washout kinetics. This is a frequent and important pitfall (discussed further in the radiologist pitfalls section below), and failing to observe this rule can lead to systematic overcategorisation of benign lesions as LR-4 or LR-5.

The hepatobiliary phase itself, acquired at approximately 20 minutes post-injection, adds substantial diagnostic value as an ancillary feature: HCC typically appears hypointense relative to background liver due to reduced or absent OATP1 expression in malignant hepatocytes, while focal nodular hyperplasia — a benign entity that can otherwise mimic HCC on earlier phases — characteristically remains iso- or hyperintense because it retains functional, if architecturally disorganised, hepatocytes and biliary drainage.

The LI-RADS treatment response algorithm (LR-TR)

A parallel and increasingly important component of the system is the LI-RADS Treatment Response algorithm (LR-TR), applied to lesions previously treated with locoregional therapy such as transarterial chemoembolisation (TACE), radiofrequency or microwave ablation, or stereotactic body radiotherapy. LR-TR categorises treated lesions as LR-TR Nonviable, LR-TR Equivocal, or LR-TR Viable, based on the presence or absence of expected treatment-related enhancement patterns versus residual nodular, mass-like APHE with washout — allowing interventional radiologists and oncologists to determine whether a treated tumour requires retreatment, all using the same multiphasic imaging data acquired under this protocol.

Reporting structure and the role of structured reporting templates

ACR-endorsed structured reporting templates operationalise LI-RADS by prompting the reporting radiologist through each major and ancillary feature systematically, reducing the risk of an omitted feature and improving inter-reader agreement — studies have demonstrated that structured LI-RADS templates measurably improve category concordance compared with free-text dictation, particularly among less experienced readers. Many PACS and voice-recognition systems now embed LI-RADS calculators directly into the reporting workflow, auto-populating the final category once the constituent features are entered, though the final categorisation always remains the responsible radiologist’s clinical judgement.

Common protocol-driven pitfalls affecting LI-RADS accuracy

Because LI-RADS categorisation depends entirely on precise phase timing and motion-free acquisition, several protocol-level failures directly propagate into reporting errors:

• Late or early arterial phase acquisition can cause a hypervascular lesion to be missed entirely (if too early, before the tumour has enhanced) or can produce false portal venous contamination that mimics washout on what is actually still an arterial-phase-like image.
• Respiratory misregistration between phases can cause a lesion to appear in a slightly different anatomical position between arterial and portal venous acquisitions, making true enhancement kinetics impossible to assess with confidence and frequently forcing a downgrade to LR-3 or an LR-NC designation.
• Incomplete hepatic coverage, particularly missing the dome of the liver in a large or elevated diaphragm, can cause a lesion to be entirely excluded from one or more phases.
• Inconsistent breath-hold depth between the pre-contrast mask and the dynamic phases degrades subtraction image quality, which many radiologists rely on as a sensitive adjunct for detecting subtle APHE.

This is the direct link between the technical protocol described earlier in this article and the diagnostic categorisation described here: a radiographer who achieves consistent navigator-gated breath-holds and a radiologist who correctly times the arterial phase are not merely following a checklist — they are the two operators most responsible for whether a given LI-RADS category is trustworthy.

Future directions

LI-RADS continues to evolve. Contrast-enhanced ultrasound LI-RADS (CEUS LI-RADS) now provides a complementary algorithm for centres with access to microbubble contrast agents, while active research is validating AI-assisted feature extraction — automated APHE and washout quantification from raw dynamic image series — as a means of further reducing inter-reader variability and flagging technically inadequate examinations before the patient leaves the department, a topic explored further in the AI and automation section of this article.

Worked example: applying the algorithm at the console

Consider a 58-year-old patient with hepatitis B-related cirrhosis undergoing surveillance liver MRI protocol imaging, in whom a 24 mm lesion is identified in segment VII. On the arterial phase, the lesion demonstrates diffuse nonrim hyperenhancement relative to background parenchyma. On the portal venous phase, the same lesion shows nonperipheral washout with a smooth, uniformly enhancing peripheral capsule not present on the pre-contrast mask. Applying the diagnostic table: size ≥20 mm plus APHE alone already meets LR-5 threshold, and the additional presence of washout and capsule further reinforces the categorisation without requiring further escalation, since LR-5 is the ceiling of HCC-specific certainty within the algorithm.

Contrast this with a second, smaller 14 mm lesion in segment IV showing APHE but no washout, no capsule, and no threshold growth on comparison with prior imaging. Under the diagnostic table, a sub-20 mm lesion with APHE as its only major feature is categorised as LR-4 rather than LR-5 — probably, rather than definitely, HCC — triggering multidisciplinary discussion rather than direct treatment planning. This size-dependent branching is precisely why accurate, reproducible lesion measurement on a motion-free acquisition is not a cosmetic reporting detail but a determinant of whether a patient proceeds straight to transplant work-up or first undergoes biopsy.

Distinguishing LR-5 from LR-M

One of the most consequential distinctions a reporting radiologist must make is between LR-5 (definite HCC) and LR-M (probably or definitely malignant, not HCC-specific), because the downstream management pathways diverge sharply — LR-5 lesions typically bypass biopsy, while LR-M lesions require histological confirmation before any treatment decision, since the differential for LR-M includes intrahepatic cholangiocarcinoma, combined hepatocellular-cholangiocarcinoma, and metastatic disease, each with a fundamentally different treatment algorithm to HCC.

CT LI-RADS versus MRI LI-RADS

Although this article focuses on the MRI-based algorithm, it is worth noting that a parallel CT LI-RADS system exists using an almost identical diagnostic table and category structure, since triphasic contrast-enhanced CT remains widely used for HCC surveillance and staging where MRI access is limited. The core major features — APHE, washout, capsule, and threshold growth — translate directly between modalities, but MRI retains several diagnostic advantages relevant to the categorisation task: superior soft-tissue contrast resolution for small (sub-centimetre) lesion detection, access to diffusion-weighted imaging as a sensitivity-boosting ancillary feature, and, where gadoxetate is used, the unique hepatobiliary phase that has no CT equivalent. For this reason, many institutional pathways reserve MRI for problem-solving of CT-indeterminate (LR-3 or LR-4) observations, or as the primary surveillance modality in patients where the added ancillary features materially change management.

Interobserver agreement and quality assurance

Multiple validation studies have assessed interobserver agreement for LI-RADS categorisation, generally reporting moderate-to-substantial agreement (kappa values in the range of 0.6–0.8) for the major features and final category assignment among experienced abdominal radiologists, with agreement measurably lower among general radiologists or trainees unfamiliar with the strict operational definitions — reinforcing the case for structured reporting templates, dedicated hepatobiliary radiologist review of indeterminate cases, and periodic departmental audit of LI-RADS category distributions against known outcomes as a quality assurance measure. Departments implementing this protocol should track the proportion of examinations designated LR-NC (technically inadequate) as a leading indicator of protocol and equipment performance, since a rising LR-NC rate typically signals a breakdown somewhere in the ten-step scanning workflow described earlier — most often at the navigator-gating or bolus-timing steps.

Integration with surveillance guidelines

LI-RADS does not exist in isolation — it is embedded within broader HCC surveillance frameworks published by bodies including the American Association for the Study of Liver Diseases (AASLD) and the European Association for the Study of the Liver (EASL), both of which recommend six-monthly surveillance imaging (typically ultrasound, with MRI or CT reserved for indeterminate or high-risk cases) in eligible cirrhotic and chronic hepatitis B populations. When a surveillance ultrasound identifies a nodule ≥10 mm, the pathway typically escalates directly to a diagnostic multiphasic CT or MRI performed under a LI-RADS-compliant protocol — meaning the dynamic liver MRI protocol described throughout this article is frequently the definitive diagnostic step in a patient’s cancer pathway, not merely a screening tool. This escalation pathway underscores why technical adequacy failures at the MRI stage carry outsized clinical consequences: a repeat scan is not merely an inconvenience but a delay in an established, guideline-mandated diagnostic timeline.

Radiology departments implementing this protocol at scale should therefore treat LI-RADS compliance as a formal quality metric alongside more traditional measures such as scanner utilisation and report turnaround time, incorporating structured feedback loops between radiographers and reporting radiologists whenever an LR-NC or equivocal category is assigned, so that recurring technical failure modes — motion, bolus mistiming, incomplete coverage — can be identified and corrected at the protocol level rather than repeatedly re-litigated on a case-by-case basis.

Top 10 pathologies

Each of the ten pathologies profiled below was selected specifically because its diagnostic confidence is materially affected by the technical quality of this protocol rather than being reliably identifiable on any single-phase or non-contrast alternative.

The dynamic multiphasic liver MRI protocol is the definitive characterisation tool for the ten focal and diffuse hepatic pathologies below. Each card lists representative T1/T2 relaxation behaviour and the specific way protocol quality influences detection or characterisation confidence. Together, these entities span the full spectrum from entirely benign (simple cyst) through indeterminate (adenoma, dysplastic nodule) to overtly malignant (HCC, cholangiocarcinoma, metastasis), and correct discrimination between them is the entire clinical purpose of performing a multiphasic rather than single-phase examination.

Note that several of these entities frequently coexist in the same patient — a cirrhotic liver may simultaneously harbour regenerative nodules, a dysplastic nodule, and an early HCC, each requiring the same multiphasic data to be distinguished from one another on a single examination. This is why per-lesion, rather than per-patient, reporting discipline is essential: every discrete observation above a reporting threshold size should be independently characterised against the major and ancillary features described in the LI-RADS section above, rather than a single overall impression being applied to the liver as a whole.

A structured approach to the indeterminate lesion

When a focal liver lesion does not fit cleanly into one of the archetypal patterns described above, a structured differential approach outperforms pattern recognition alone. The first branch point is arterial phase behaviour: a lesion showing nonrim APHE in a cirrhotic liver should be evaluated against the LI-RADS major features described earlier, whereas a hypervascular lesion in a non-cirrhotic liver more commonly represents FNH, adenoma, or a hypervascular metastasis (renal cell carcinoma, neuroendocrine tumour, thyroid carcinoma, melanoma), reinforcing why clinical risk-factor context must always accompany image interpretation rather than being assessed in isolation.

The second branch point is enhancement pattern morphology: peripheral nodular discontinuous enhancement with progressive centripetal fill-in is essentially pathognomonic for haemangioma across virtually all lesion sizes, while progressive delayed peripheral-to-central enhancement — the inverse temporal pattern — is the hallmark of cholangiocarcinoma and reflects its fibrous, hypocellular stroma. The third branch point, particularly relevant when the first two are equivocal, is hepatobiliary-phase behaviour on gadoxetate-enhanced examinations, which frequently resolves the FNH-versus-adenoma-versus-well-differentiated-HCC differential that arterial and portal venous phases alone cannot.

Diffusion-weighted imaging contributes a fourth, largely independent axis of information: most malignant lesions, including HCC and metastases, show restricted diffusion (low apparent diffusion coefficient) reflecting high cellularity, while haemangiomas and simple cysts characteristically show facilitated diffusion (high ADC) due to their fluid content, providing a valuable cross-check when contrast-based features are ambiguous or when a study is technically limited due to motion.

Pitfalls — radiographers

The primary scanning pitfall identified for this protocol is respiratory motion blur, occurring when breath-hold depth varies between the pre-contrast mask and the dynamic phases, or when a patient takes an involuntary breath during a nominally “held” acquisition — most catastrophically during the irrecoverable arterial phase.

Respiratory motion blur is disproportionately consequential in this protocol compared with most other body MRI examinations because the arterial phase cannot be repeated: once the contrast bolus has redistributed into the portal venous and equilibrium pools, the specific 5–8 second window of pure arterial enhancement is gone for that injection. A motion-degraded portal venous or delayed phase, while undesirable, can sometimes be repeated using a small supplemental contrast dose or accepted with a caveat; a motion-degraded arterial phase in a cirrhotic surveillance patient frequently necessitates rescheduling the entire examination on a separate day with a fresh contrast dose, at real cost to the patient, the department, and the diagnostic pathway timeline discussed in the LI-RADS surveillance section above.

Pitfalls — radiologists

The primary interpretation pitfall for this protocol is misreading gadoxetate-related pseudo-washout on the transitional phase as true HCC washout, leading to systematic overcategorisation of benign lesions.

This pitfall arises specifically because gadoxetate-enhanced examinations include a transitional phase (roughly 2–5 minutes post-injection) during which background hepatocytes are actively beginning to take up contrast via OATP1 transporters ahead of the formal hepatobiliary phase. A radiologist accustomed to extracellular-agent protocols, where any phase after the portal venous phase can be reasonably used to assess washout, may inadvertently apply that same logic to a gadoxetate examination and interpret the relative hypointensity of a benign lesion against increasingly enhancing background liver as true malignant washout — a systematic bias that the LI-RADS v2018 technical requirement to assess washout strictly on the portal venous phase was specifically written to prevent.

A common thread across all four radiologist-facing pitfalls above is the temptation to interpret a single phase in isolation rather than as part of a continuous kinetic story. Confident LI-RADS categorisation depends on evaluating the full time-course of enhancement across all acquired phases together with the pre-contrast mask and subtraction series, rather than anchoring on any one image, a discipline that becomes considerably harder, and considerably more important, when the underlying acquisition is technically imperfect.

Pitfalls — non-radiology physicians

Referring clinicians who order and act on this protocol without a radiology background face a distinct set of pitfalls, centred not on image acquisition or interpretation but on correctly translating a structured LI-RADS category into an appropriate clinical action. Because LI-RADS categories map to specific, guideline-defined management pathways rather than a simple benign/malignant binary, a surface-level reading of the report can lead a well-intentioned clinician toward either false reassurance or unnecessary intervention.

Pitfall comparison summary

The three pitfall domains above are not independent — a scanning-stage failure frequently cascades directly into an interpretation-stage or clinical-stage error, since a radiologist working with motion-degraded images is more likely to render an equivocal or incorrect category, which in turn is more likely to be misapplied downstream. Viewing the pitfall framework side-by-side makes this cascade visible and helps departments identify where a single upstream intervention — such as consistent navigator gating — can prevent errors from propagating through the entire diagnostic chain.

Read together, these three columns describe a single continuous quality chain rather than three unrelated checklists. A department auditing its liver MRI performance should track indicators at each stage — LR-NC rate for scanning quality, category concordance on second review for interpretation quality, and follow-up compliance rates for clinical-stage quality — since each metric illuminates a different point in the chain where the same underlying protocol can succeed or fail.

AI and automation

Artificial intelligence is increasingly embedded across the dynamic liver MRI protocol workflow, from acquisition through to reporting. Deep-learning image reconstruction tools (such as vendor-neutral and vendor-specific denoising reconstruction algorithms, several of which carry FDA clearance and CE marking) allow shorter breath-hold acquisitions at equivalent image quality by permitting higher parallel-imaging acceleration factors without a proportional SNR penalty — directly reducing the window in which a patient must remain motionless.

Beyond reconstruction, several vendors now offer AI-driven respiratory motion prediction, which learns an individual patient’s breathing pattern during the pre-contrast localiser and mask acquisition and uses that learned pattern to optimise trigger timing for the subsequent contrast-enhanced phases, a direct technological response to the respiratory motion blur pitfall identified as the primary scanning failure mode for this protocol. Early clinical validation of these tools reports meaningful reductions in non-diagnostic arterial phase rates compared with standard navigator gating alone, though prospective multicentre confirmation remains ongoing.

Automated bolus-tracking software, integrated with MR-compatible power injectors, standardises arterial-phase timing across operators and reduces reliance on individual radiographer judgement. On the interpretation side, AI-assisted lesion detection and LI-RADS feature-extraction algorithms are under active clinical validation, with published multicentre studies demonstrating automated systems performing comparably to mid-level radiologists^[9] for focal liver lesion detection and classification on multiphase imaging, and improving the diagnostic accuracy of junior radiologists when used as a second reader.^[9]

Regulatory clearance status matters when evaluating any AI tool for clinical deployment. FDA 510(k) clearance or CE marking under the EU Medical Device Regulation indicates a defined intended use, validated performance claims, and post-market surveillance obligations; departments should verify current clearance status directly with the vendor and relevant regulatory database before incorporating any AI-assisted reconstruction, triggering, or reporting tool into routine clinical workflow, since clearance status and approved indications can change over time.

Further reading

The following resources expand on specific technical themes covered in this article, from precision contrast delivery and gadoxetate-related motion artefact to sibling protocols in this series that illustrate analogous timing-critical and motion-sensitive acquisition principles.

1. 7 Proven Strategies for Optimizing MRI Sequences in 2026 — covers gadoxetate-related transient severe motion and precision contrast delivery in depth.
2. Gadolinium-Enhanced MRI: Enhancement Patterns, Imaging Protocols, and Optimal Contrast Timing — a literature review of contrast-timing principles applicable across dynamic multiphasic protocols.
3. Acute Stroke MRI Protocol — a sibling protocol in this series illustrating time-critical, motion-sensitive MRI acquisition principles.
4. CT Trauma Pan-Scan Protocol — demonstrates multiphase contrast timing principles analogous to hepatic arterial/portal venous phase separation.
5. 7 Critical CTA Brain & Carotids Protocol Steps Every Radiographer Must Master — covers high-pressure injector technique and bolus-tracking principles relevant to power injection for dynamic MRI.

Reducing artefacts: patients and parameters

While the earlier sections of this article addressed protocol-specific mitigation strategies such as navigator gating and bolus timing, the underlying MRI physics parameters that govern spatial resolution, signal-to-noise ratio, contrast, and artefact behaviour apply universally across every sequence in this examination. A radiographer who understands these fundamental trade-offs is equipped to make sound real-time adjustments when a patient cannot tolerate the standard protocol — shortening a breath-hold, accepting a modest resolution reduction, or switching to a free-breathing fallback — without simply abandoning diagnostic quality altogether.

The most critical scanning parameters that impact image quality in the dynamic liver MRI protocol fall into four interlinked domains, each with direct trade-offs the operator must balance in real time.

Spatial resolution

Spatial resolution defines the ability to distinguish small details in an image. Matrix size: increasing the matrix (frequency × phase) increases spatial resolution but decreases SNR because the voxel size becomes smaller. Field of view (FOV): reducing FOV increases spatial resolution, though smaller voxels again reduce SNR. Slice thickness: thinner slices provide higher spatial resolution and reduce partial volume averaging but significantly decrease SNR — a trade-off that is especially consequential when trying to resolve sub-centimetre HCC nodules against background liver.

Signal-to-noise ratio (SNR)

SNR represents the strength of the diagnostic signal relative to inherent background noise. A high SNR produces crisp, clear images, whereas a low SNR looks grainy. Number of averages (NEX/NSA): increasing averages improves SNR by acquiring data multiple times, but doubling averages roughly doubles scan time — rarely feasible within a single breath-hold. Receiver bandwidth: decreasing bandwidth limits recorded noise and boosts SNR, but lowers scan speed and increases chemical shift artefact. Coil selection: dedicated, localised surface/torso phased-array coils capture substantially stronger signal than whole-body coils and heavily improve SNR.

Image contrast

Contrast determines how different tissues are distinguished from one another. Repetition time (TR): the time between consecutive RF pulses; a short TR maximises T1 tissue contrast (essential for the dynamic phases), while a long TR minimises it. Echo time (TE): the time between the RF pulse and peak echo signal; a short TE minimises T2 effects, while a long TE maximises T2 weighting, making fluid-filled structures appear very bright. Flip angle: controls proton excitation and directly changes tissue contrast, particularly critical in the gradient-echo sequences that underpin this entire protocol.

Artefact control

Artefacts are visual distortions or ghosting that degrade image quality. Phase-encoding direction: swapping phase and frequency axes can shift motion-induced artefacts — such as breathing or vascular flow — away from the liver parenchyma and lesion of interest. Flow compensation / gating: utilises physiological triggers (respiratory navigator, cardiac gating) to minimise blurring and ghosting from pulsatile or respiratory motion — the single most important lever for this protocol. Parallel imaging: uses multiple coil elements simultaneously to reduce the number of phase-encoding steps required, significantly cutting scan time and, by extension, reducing the opportunity for motion to occur within any one acquisition.

Parallel imaging protocols and parameters

Parallel imaging acceleration is central to compressing each dynamic phase into a breath-holdable window. Turbo/acceleration factor selection must balance scan-time reduction against the SNR penalty inherent to undersampled k-space reconstruction.

The relationship between acceleration factor and image quality is not linear: each incremental increase in acceleration factor reduces scan time roughly proportionally but increases noise amplification (the “g-factor”) in a manner that depends heavily on coil geometry and the anatomical region being imaged. In the liver, where the region of interest sits centrally within the torso coil array rather than near individual coil elements, g-factor penalties tend to be higher than in more peripheral anatomy such as the shoulder or knee, meaning acceleration factors that work well elsewhere in the body may require adjustment for acceptable liver image quality.

Compressed sensing, which exploits the inherent sparsity of MR images in an appropriate transform domain rather than relying solely on coil sensitivity encoding, has become an increasingly important complementary acceleration technique for this protocol. When combined with conventional parallel imaging, compressed sensing can permit total acceleration factors well beyond what either technique achieves alone, translating directly into shorter, more achievable breath-holds — a meaningful advantage in the elderly, dyspnoeic, or post-surgical patient populations who make up a disproportionate share of liver MRI referrals.

Conclusion

Radiology departments implementing or auditing this protocol should treat the guidance above as a practical checklist: verify navigator or respiratory-gating hardware is functioning and correctly positioned before every list; confirm bolus-timing technique (fluoroscopic trigger or timing bolus) is used for every patient rather than a fixed empirical delay; ensure reporting radiologists apply LI-RADS washout assessment strictly to the correct phase for the agent in use; and confirm referring clinicians receive clear, category-specific guidance rather than an ambiguous free-text impression.

The dynamic multiphasic liver MRI protocol demands a level of coordinated precision — between radiographer, injector, and patient physiology — matched by few other examinations in diagnostic imaging. Success rests on three interlocking pillars: technically sound acquisition built around navigator-gated breath-hold consistency and accurately timed arterial-phase bolus delivery; a rigorous understanding of hepatic anatomy, relaxation behaviour, and the ten pathologies most commonly characterised on this protocol; and disciplined application of the LI-RADS framework, which converts precisely acquired imaging into a reproducible, actionable category with direct management consequences.

The pitfall framework outlined across this article — respiratory motion blur at the scanning stage, gadoxetate pseudo-washout at the interpretation stage, and misapplied or misunderstood categorisation at the clinical stage — represents the three points in the diagnostic chain where protocol quality most directly determines patient outcome. Departments that systematically address each of these three failure points, through equipment, training, and structured reporting, will consistently produce the motion-free, correctly timed, LI-RADS-compliant examinations on which confident hepatic oncology decision-making depends.

No single intervention solves the liver MRI protocol’s challenges in isolation. Reliable navigator-gated breath-hold acquisition, reproducible low-rate contrast delivery, structured LI-RADS reporting, and clear communication of category-specific management pathways to referring clinicians together form a chain that is only as strong as its weakest link. As deep-learning reconstruction, AI-assisted motion prediction, and automated feature extraction continue to mature, the technical burden on any single operator will likely decrease — but the underlying physiological reality that the liver moves with every breath, and that the arterial phase can be acquired only once, will remain the defining constraint of this protocol for the foreseeable future.

This checklist-driven approach, applied consistently across every examination, is what separates a department that occasionally produces excellent liver MRI studies from one that reliably does so on every patient, every time.

References

1. American College of Radiology. (2018). CT/MRI Liver Imaging Reporting and Data System (LI-RADS) version 2018. American College of Radiology. https://www.acr.org/Clinical-Resources/Reporting-and-Data-Systems/LI-RADS
2. Marrero, J. A., Kulik, L. M., Sirlin, C. B., Zhu, A. X., Finn, R. S., Abecassis, M. M., Roberts, L. R., & Heimbach, J. K. (2018). Diagnosis, staging, and management of hepatocellular carcinoma: 2018 practice guidance by the American Association for the Study of Liver Diseases. Hepatology, 68(2), 723–750. https://doi.org/10.1002/hep.29913
3. European Association for the Study of the Liver. (2018). EASL clinical practice guidelines: Management of hepatocellular carcinoma. Journal of Hepatology, 69(1), 182–236. https://doi.org/10.1016/j.jhep.2018.03.019
4. Chernyak, V., Fowler, K. J., Kamaya, A., Kielar, A. Z., Elsayes, K. M., Bashir, M. R., Kono, Y., Do, R. K., Mitchell, D. G., Singal, A. G., Tang, A., & Sirlin, C. B. (2018). Liver Imaging Reporting and Data System (LI-RADS) version 2018: Imaging of hepatocellular carcinoma in at-risk patients. Radiology, 289(3), 816–830. https://doi.org/10.1148/radiol.2018181494
5. Davenport, M. S., Malyarenko, D. I., Pang, Y., Hussain, H. K., & Chenevert, T. L. (2018). Effect of gadoxetate dilution and slow injection rate on the incidence of transient severe motion artifact in gadoxetate disodium-enhanced MRI of the liver. American Journal of Roentgenology, 210(3), 561–566. https://doi.org/10.2214/AJR.17.18581
6. Bashir, M. R., Castelli, P., Davenport, M. S., Larson, D., Marin, D., Hussain, H. K., & Semelka, R. C. (2015). Respiratory motion artifact affecting hepatic arterial phase MR imaging in patients receiving gadoxetate disodium: Frequency, workflow implications, and patient risk factors. Radiology, 274(2), 405–415. https://doi.org/10.1148/radiol.14140690
7. Motosugi, U., Bannas, P., Bookwalter, C. A., Sano, K., & Reeder, S. B. (2016). An investigation of transient severe motion related to gadoxetic acid-enhanced MR imaging. Radiology, 279(1), 93–102. https://doi.org/10.1148/radiol.2015150642
8. Zech, C. J., Bartolozzi, C., Bioulac-Sage, P., Chow, P. K., Forner, A., Grazioli, L., Kudo, M., & Reeder, S. B. (2020). Consensus report from the 8th International Forum for Liver Magnetic Resonance Imaging. European Radiology, 30(1), 370–382. https://doi.org/10.1007/s00330-019-06369-4
9. Wang, K., Zhang, X., Zhang, B., Deng, J., Zhang, L., Guo, Y., Cui, Y., Wu, D., & Yang, X. (2024). A multicenter clinical AI system study for detection and diagnosis of focal liver lesions. Nature Communications, 15, 1131. https://doi.org/10.1038/s41467-024-45325-9
10. Ronot, M., Fouque, O., Esvan, M., Lebigot, J., Aubé, C., & Vilgrain, V. (2018). Comparison of the accuracy of AASLD and LI-RADS criteria for the non-invasive diagnosis of HCC smaller than 3 cm. Journal of Hepatology, 68(4), 715–723. https://doi.org/10.1016/j.jhep.2017.11.030
11. Kim, T. H., Woo, S., Han, S., Suh, C. H., Vietti-Violi, N., & Taouli, B. (2020). The diagnostic performance of LI-RADS treatment response algorithm for hepatocellular carcinoma: A systematic review and meta-analysis. Radiology, 296(2), 336–345. https://doi.org/10.1148/radiol.2020192785
12. International Commission on Radiological Protection. (2016). ICRP Publication 132: Radiological protection and safety in medical uses of ionizing radiation and non-ionizing fields. ICRP.
13. International Electrotechnical Commission. (2015). IEC 60601-2-33: Medical electrical equipment — Particular requirements for the basic safety and essential performance of magnetic resonance equipment for medical diagnosis (Ed. 3.2). IEC.
14. American Association of Physicists in Medicine. (2019). AAPM report: MR safety and specific absorption rate management practice guidelines. AAPM.
15. European Commission. (2016). EC Radiation Protection No. 185: Medical radiation exposure of the European population. Publications Office of the European Union.
16. Tamada, D. (2020). Breathing artifacts in liver MRI: Emerging solutions. Proceedings of the International Society for Magnetic Resonance in Medicine (ISMRM), Abstract E875.
17. Wu, W., Chen, X., Yang, Y., Zhu, K., Cui, C., & Ye, J. (2022). Free-breathing liver magnetic resonance imaging with respiratory frequency-modulated continuous-wave radar-trigger technique: A preliminary study. Frontiers in Oncology, 12, 918173. https://doi.org/10.3389/fonc.2022.918173
18. American College of Radiology. (2023). ACR Manual on Contrast Media, version 2023. American College of Radiology.
19. Woo, S., Lee, J. M., Yoon, J. H., Joo, I., Han, J. K., & Choi, B. I. (2015). Comparison of the diagnostic performance of gadoxetic acid-enhanced MRI and gadopentetate-enhanced MRI for the detection of hepatocellular carcinoma. Radiology, 276(3), 792–801. https://doi.org/10.1148/radiol.2015141315
20. Fowler, K. J., Tang, A., Santillan, C., Bhargavan-Chatfield, M., Heiken, J., Jha, R. C., Fetzer, D., Kono, Y., Kamaya, A., Sirlin, C. B., & Chernyak, V. (2018). Interreader reliability of LI-RADS version 2014 algorithm and imaging features for diagnosis of hepatocellular carcinoma. Radiology, 286(1), 173–185. https://doi.org/10.1148/radiol.2017170114
21. Elsayes, K. M., Kielar, A. Z., Elmohr, M. M., Chernyak, V., Masch, W. R., Furlan, A., Marks, R. M., Bashir, M. R., Jambhekar, K., Fowler, K. J., & Heiken, J. P. (2018). White paper of the Society of Abdominal Radiology hepatocellular carcinoma diagnosis disease-focused panel on LI-RADS v2018 for CT and MRI. Abdominal Radiology, 43(11), 2625–2642. https://doi.org/10.1007/s00261-018-1744-4
22. Van der Pol, C. B., Lim, C. S., Sirlin, C. B., McGrath, T. A., Salameh, J. P., Bashir, M. R., Tang, A., Singal, A. G., Costa, A. F., Fowler, K., & McInnes, M. D. F. (2019). Accuracy of the Liver Imaging Reporting and Data System in computed tomography and magnetic resonance image analysis of hepatocellular carcinoma: A systematic review. Gastroenterology, 156(4), 976–986. https://doi.org/10.1053/j.gastro.2018.11.020
23. Chartampilas, E., Rafailidis, V., Georgopoulou, V., Kalarakis, G., Hatzidakis, A., & Prassopoulos, P. (2022). Current imaging diagnosis of hepatocellular carcinoma. Cancers, 14(16), 3997. https://doi.org/10.3390/cancers14163997
24. Kim, B. S., Bae, J. S., & Kim, J. H. (2019). Robust liver MR sequences to image patients with poor breath-hold capabilities. BioMed Research International, 2019, 1780205. https://doi.org/10.1155/2019/1780205
25. Chow, P. K., Choo, S. P., Ng, D. C., Lo, R. H., Wang, M. L., Toh, H. C., Tai, D. W., Goh, B. K., Wong, J. S., Tay, K. H., & Chung, A. Y. (2018). National Cancer Centre Singapore consensus guidelines for hepatocellular carcinoma. Liver Cancer, 8(6), 453–472. https://doi.org/10.1159/000496167
26. Cannella, R., Jang, H. J., Borhani, A. A., Rachiwong, S., Rosado, E., Furlan, A., & Dioguardi Burgio, M. (2023). Evolution of the LI-RADS treatment response algorithm: Diagnostic performance and future directions. Insights into Imaging, 14, 40. https://doi.org/10.1186/s13244-023-01380-1
27. Ehman, E. C., Wells, M. L., Fidler, J. L., Nagayama, Y., Trout, A. T., Dane, B., Wang, C., Kambadakone, A., Fung, A., Elsayes, K. M., & Bashir, M. R. (2022). Predicting outcomes of small (10–19 mm) indeterminate observations on CT and MRI in at-risk patients using LI-RADS version 2018. Abdominal Radiology, 47(2), 566–577. https://doi.org/10.1007/s00261-021-03337-7
28. Choi, S. H., Byun, J. H., Lim, Y. S., Yu, E., & Lee, S. J. (2019). Diagnostic criteria for hepatocellular carcinoma ≤3 cm with hepatocyte-specific contrast-enhanced magnetic resonance imaging. Journal of Hepatology, 66(6), 1130–1137. https://doi.org/10.1016/j.jhep.2017.01.028