MRI Scanner Monthly ACR Phantom Quality Assurance & Performance Log

ContextThe phantom must be centered within ±5mm in all three axes to ensure the prescribed axial slices pass through the correct internal structures. Use the foam positioning aids supplied with the phantom kit to prevent roll or lateral shift during table travel. Miscentering by even 10mm laterally can cause the geometric accuracy measurement slices to sample non-representative sections of the phantom, either inflating or masking true distortion values. Confirm the flat side of the phantom (which contains the slice position accuracy wedges) is at the head end of the table — this is the ACR convention, and reversing the orientation will cause incorrect scoring on Test 4 without any other visible indication of error.

ContextMonthly ACR testing must be performed with the head coil because the ACR pass criteria were derived from phantom data acquired with this coil. Substituting any other coil — including a flex array or spine coil — produces systematically different SNR and uniformity values that cannot be compared to ACR thresholds. Before scanning, verify the coil element status display: if one or more elements show as disconnected or shorted, your uniformity and SNR measurements will be invalid and any pass result will be meaningless. Note the exact coil model and serial number in the log; if the facility has multiple head coils, always use the same unit for serial trending unless you intentionally re-baseline the control charts after a coil swap.

ContextRecording the table entry position at isocenter is a small but critical step. If the scanner is serviced and the table zero reference is recalibrated, comparing monthly entry values can reveal the shift before it propagates to clinical positioning errors in patient exams. Some facilities formalize this by requiring the entry position to be within ±2mm of the recorded baseline as an independent QA pass criterion. After locking, run a quick laser alignment check to confirm no creep occurred — some older motorized table drives exhibit a small position drift in the seconds following the lock command.

ContextACR accreditation requires QA images to be acquired using the exact protocol parameters specified for 1.5T or 3T phantom testing: flip angle, TR, TE, FOV 250mm, slice thickness 5mm, matrix 256×256, receive bandwidth, and phase-encode direction must all match the ACR specification. Modifying any parameter — including seemingly minor choices like phase encoding direction or readout bandwidth — can shift multiple test results outside their pass thresholds while the images look visually normal. If the protocol file has been altered since the last run, contact your medical physicist to restore it before proceeding. Some scanners auto-prompt for parameter adjustments during loading; decline any prompt and investigate the cause before scanning.

ContextThe localizer is your sanity check before committing to the full diagnostic series. The phantom should appear as a clean rectangle on sagittal and coronal localizers, and as a near-perfect circle on the axial — any skew greater than approximately 2° indicates the phantom is not orthogonally positioned and all subsequent tests will be compromised. Stop, reposition, and reacquire before proceeding. On some 3T systems, the localizer will also reveal shim-related banding artifacts if the active shim coils are malfunctioning; identifying this on the localizer allows you to perform a manual shim before the T1 and T2 series rather than discovering the problem after full acquisition.

ContextThe T1 axial series is used for Tests 1 through 5: geometric accuracy, high-contrast resolution, slice thickness accuracy, slice position accuracy, and percent image uniformity. Missing even one slice invalidates multiple tests simultaneously. After acquisition, scroll through the complete series on the console and check for three specific artifact patterns: banding across slices (indicating B1 non-uniformity or a dielectric effect at 3T), ghosting propagating through the phantom in the phase-encode direction (indicating motion or gradient instability), and abrupt intensity steps between consecutive slices (indicating a multi-channel coil element combining error in reconstruction).

ContextThe T2 series is used exclusively for Test 7 (low contrast object detectability), the most sensitive indicator of scanner performance degradation in the entire ACR protocol. At 3T, the T2 series is particularly vulnerable to B0 field drift if the magnet has not fully stabilized after a recent disturbance. Susceptibility artifacts near the phantom wall on slices 1–4 — which contain the low-contrast disk inserts — can obscure small objects and produce a false failure result. If these artifacts appear, document their presence, then perform a manual shim adjustment and reacquire; if they clear after shimming, note the required shim values and flag the finding for physicist review as an indicator of field stability issues.

ContextThe ACR phantom has a known inner diameter of 190mm. The ACR pass criterion requires all measurements to be within ±2mm of this known value, giving a pass window of 188–192mm in every measured direction. Geometric distortion is caused by gradient nonlinearity, eddy currents, B0 field non-uniformity, or gradient calibration drift. A consistent undershoot of 1–2mm in one specific direction suggests gradient scaling error; measurements that vary erratically from session to session suggest eddy current instability rather than a static calibration offset. Also measure the long axis of the phantom on a sagittal reconstruction — the known axial length is 148mm — to capture distortion along the bore axis, which gradient scaling errors affect independently of the transverse axes.

ContextDistortion percentage = ((measured value − known value) / known value) × 100. Record all six measurements: axial horizontal, axial vertical, sagittal long axis, sagittal width, coronal long axis, and coronal width. Even when all values pass the ±2mm threshold, record actual numbers rather than checkmarks — the numerical record is the trend data. A scanner that passes at 1.8mm distortion for six consecutive months and then jumps to 1.9mm warrants a physicist notification even though it technically still passes, because the direction of change predicts a future threshold crossing and the cost of catching it early is a phone call versus an emergency service call.

ContextEach array consists of a 4×4 grid of cylindrical holes drilled to a specific diameter. An array is considered resolved if you can clearly distinguish all four holes in any 2×2 subgroup as separate structures, without partial-volume blurring merging adjacent holes into a single signal void. The ACR minimum pass criterion is that the 1.0mm array must be resolved in both the frequency-encode and phase-encode directions. Many facilities set an internal performance target of the 0.9mm array to provide a margin above the minimum threshold. Degradation in this test can indicate RF receiver bandwidth narrowing, gradient calibration drift affecting the readout axis, or a subtle change in the reconstruction anti-aliasing filter coefficients.

ContextResolution is frequently asymmetric between the two in-plane directions — the frequency-encode axis is governed by gradient bandwidth and ADC sampling rate, while the phase-encode axis depends on k-space sampling density and any phase correction applied during reconstruction. Passing in one orientation but failing in the other is diagnostically meaningful and points to different root causes. Phase-encode resolution failure that developed after a software update often indicates that a reconstruction filter has become more aggressive in that axis. Always record both results even when both pass, because asymmetry that develops over time is itself a trend worth tracking.

ContextThe ACR phantom contains two crossed ramps made of signal-producing material, each mounted at a precise ±10° angle from the imaging plane. When a 5mm slice intersects these ramps, each produces a signal line whose projected length encodes the actual slice thickness. Calculate slice thickness using: ST = (top ramp length × bottom ramp length) / (top ramp length + bottom ramp length). The ACR pass criterion is a calculated slice thickness within ±1mm of the prescribed 5mm, establishing a pass window of 4.0mm to 6.0mm. Slice thickness errors trace back to RF pulse miscalibration, gradient slice-select amplitude errors, or eddy currents on the slice-select gradient axis.

ContextThe two ramps serve as redundancy checks on each other. If they disagree by more than 0.5mm, the slice profile is asymmetric — common causes include an air bubble partially obscuring one ramp, RF pulse asymmetry due to transmitter calibration drift, or a DC offset on the slice-select gradient. Asymmetry of this magnitude should be noted in the log even if the averaged value still passes. Tracking the trend of individual ramp lengths over months is a sensitive early warning: a slice thickness creeping from 5.0mm toward 5.8mm over six months is a classical signature of progressive slice-select gradient scaling drift that will eventually fail the test if not addressed.

ContextThe ACR phantom contains paired wedge-shaped signal inserts near each flat end. When a slice lands at its prescribed position, the two bars of each wedge pair appear equal in length. Displacement from the prescribed position causes one bar to lengthen and the other to shorten. Calculate position offset as: offset (mm) = 5 × (top bar length − bottom bar length) / (top bar length + bottom bar length). Measure both slices 1 and 11 because they straddle the full prescribed range of the phantom and are the most sensitive to slice registration errors — they will detect table positioning drift, gradient calibration offset, and isocenter definition errors before the middle slices do.

ContextThe sign of the offset carries diagnostic information: a positive value means the slice is displaced superiorly, a negative value means inferiorly. A scanner that consistently measures at −3mm is differently calibrated than one that varies randomly around zero — the former indicates a systematic isocenter definition offset while the latter suggests mechanical or electronic instability. The ACR pass threshold is ±5mm, but many academic and high-volume centers use ±3mm as a tighter internal threshold to provide warning before the ACR limit is reached. When slices 1 and 11 both fail in the same direction simultaneously, the most likely cause is a table position error rather than a gradient problem.

ContextSlice 7 is the designated uniformity measurement slice because it falls in the central portion of the phantom, free of internal structures, providing a homogeneous signal target with no contamination from internal phantom features. The large ROI is required — using a small ROI misses regional non-uniformities that a large ROI would capture and could cause a failing scanner to appear to pass. Most PACS workstations and QA tools report the maximum and minimum pixel values within the ROI automatically. If calculating manually, scan a small sub-ROI (approximately 1 cm²) systematically across the entire phantom interior to find the true global maximum and minimum pixel intensities.

ContextThe ACR minimum threshold is 87.5%. At 3T, non-uniformity is a persistent challenge due to dielectric resonance effects (B1 standing waves) that create a central bright spot flanked by darker peripheral zones — this is a physics phenomenon, not a scanner malfunction, but it can push marginal units below the threshold. If your facility uses dielectric pads or RF shimming for clinical brain imaging, remove them for QA to capture the scanner's unaugmented baseline. Uniformity values between 87.5% and 90% should be flagged for active monitoring; a downward trend of more than 2 percentage points between consecutive monthly sessions warrants physicist notification. RF coil element dropout, preamplifier failures, and image filter changes all manifest most prominently in this test.

ContextGhost artifacts propagate only along the phase-encode direction, so the four cardinal ROIs must be positioned on the phase-encode axis — not arbitrarily at the four compass points of the image. If you are unsure of the phase-encode direction for this acquisition, check the scan parameters before placing ROIs. Also place a fifth background ROI in a corner of the image (as far from the phantom as possible, away from any visible ghost streaks) to measure the ambient noise floor. This background ROI is essential: without it you cannot calculate PSG, only observe it qualitatively.

ContextThis formula normalizes the ghost signal intensity against the phantom's own signal, yielding a dimensionless percentage. A PSG above 2.5% means a real-image artifact is being superimposed on clinical anatomy in the ghost propagation zone — a direct patient safety concern for any brain or spine study where small lesions could be masked or mimicked by a signal shadow. Common causes of PSG failure include a loose or intermittently contacting coil cable, vibration from nearby HVAC compressors or construction transmitted through the floor slab, a recently installed gradient shim that introduced a new mechanical resonant frequency, or a vendor software update that altered the phase correction algorithm applied during image reconstruction.

ContextEach of the four slices contains multiple spokes radiating from the center of the phantom, with each spoke composed of a series of cylindrical plugs at progressively lower contrast levels relative to the background (spanning approximately 1% to 6% signal difference). A spoke counts toward the score only if every plug in that spoke is individually visible — a single unresolved plug invalidates the entire spoke. This is the most subjective of the seven tests and carries the highest inter-observer variability. The ACR requires that the same trained observer consistently perform the scoring session to maintain valid trending, or that multiple observers score independently and document their agreement rate.

ContextThe pass threshold differs significantly by field strength: the 1.5T minimum requires at least 9 total spokes detected (with specific per-slice distribution requirements), while the 3T minimum requires at least 37 total objects detectable across the designated slices. The applicable threshold must be documented in your QA template and communicated clearly to every observer who will score this test — applying the wrong threshold has caused both false failures and false passes at facilities that upgraded field strength but did not update their QA forms. Never attribute a LCOD failure to observer variability without having a second qualified observer independently rescore from the same images before drawing that conclusion.

ContextSNR measurement by the single-image method (aligned with NEMA MS1-2008) measures mean signal in the phantom center and noise in the same image background — faster than the two-acquisition subtraction method and sufficient for monthly trending purposes. Centering the ROI at least 10mm from the phantom wall prevents the partial-volume effect from the phantom casing material contaminating the signal measurement. Crucially, use the identical ROI size and center position every month — even a small shift toward the phantom edge can introduce a reproducible positive or negative bias in S, creating the appearance of a trend that does not reflect real scanner performance change.

ContextThe scaling factor 0.655 is a Rician correction for the noise statistics of magnitude MR images. In magnitude reconstruction, random noise adds in quadrature rather than averaging to zero, which elevates the mean background signal above the true noise floor — without correction, single-image SNR is overestimated by approximately 52%. The background ROI must be placed far from the phantom boundary and from any visible ghosting artifact in the phase-encode direction. Record the calculated SNR value on your control chart. A 10% decline from the established baseline is the standard physicist notification threshold; a 20% decline constitutes a hard failure that should immediately suspend clinical scanning on that unit pending investigation and service.

ContextStatistical process control — specifically Shewhart control chart rules applied to the SNR data series — is significantly more sensitive than simple static pass/fail thresholds for detecting real performance changes early. Track the rolling mean and standard deviation of the SNR dataset and apply the following rules: flag any single point beyond 3σ from the mean (indicating a sudden discrete event), any two consecutive points beyond 2σ (indicating persistent degradation), or any six consecutive points trending monotonically in one direction (indicating slow drift). Sudden drops suggest discrete hardware events such as a coil preamplifier failure or a new RF interference source. Slow monotonic declines suggest consumable aging such as coil cable shield degradation or RF amplifier gain drift over time.

ContextACR accreditation reviewers expect a complete numerical record — not checkboxes. If you recorded only a pass result for geometric accuracy but the actual measurements were 189.5mm and 190.2mm, you have discarded the trend data needed to predict that a future session will fail at 187.8mm. Each measurement should be recorded to the nearest 0.1mm for linear dimensions and 0.1% for ratio metrics. The log should additionally include the scanner's host software version and table drive firmware version, because vendors occasionally push automatic background updates that alter scanner behavior without generating a formal service event notification visible in the service log.

ContextTrend analysis is the primary value of monthly QA compared to annual accreditation testing: a single data point tells you whether you passed or failed today, while twelve data points tell you where the scanner is heading over the next year. The three most important trend patterns to identify are: a monotonic drift toward the failure threshold (predicts failure), a sudden single-session step change (suggests a discrete service or software event), and increasing month-to-month variability (suggests an intermittent hardware fault that behaves differently depending on environmental or load conditions on the day of testing). Automated QA platforms such as Magnet-Assist or ACR QA Suite generate these charts automatically; manual spreadsheets should use conditional color formatting to highlight out-of-trend cells.

ContextThe margin between current performance and the failure threshold is as operationally important as the pass/fail result itself. A scanner at PIU 88.5% against a threshold of 87.5% has a one-point safety margin — a single bad session could fail it and trigger the entire corrective action and clinical suspension process. Formally annotating watch-status parameters creates a documented record showing that the facility was actively monitoring a marginal parameter rather than discovering a failure unexpectedly. This documentation is specifically evaluated by ACR site reviewers and Joint Commission surveyors as evidence of a proactive, functioning QA program rather than a reactive one.

ContextACR accreditation requires QA images to be retained and available for review for the duration of the accreditation cycle, typically three years. Storing images only on the scanner's local console hard drive is insufficient: drive replacements during service events routinely result in complete data loss that is not recoverable. A PACS archive with a consistent, systematic naming convention allows any accreditation reviewer, site physicist, or compliance officer to locate a specific month's images within seconds. Never archive QA images as JPEG screenshots or bitmap exports — these formats strip the DICOM header metadata that records acquisition parameters, and without that metadata the images cannot be verified as having been acquired with the required protocol.

ContextACR accreditation standards require that a qualified medical physicist (QMP) review and approve any corrective action plan before a failed scanner returns to routine clinical use. For facilities with a staff physicist, call directly and send the full QA packet with images. For facilities using a consulting physicist, send the complete packet via secure email immediately and document the time of transmission. If clinical studies were performed during the period when the QA failure was present (identified retrospectively), the QMP must evaluate whether those study results are diagnostically valid and determine whether the ordering physicians need to be notified — this is a formal regulatory requirement, not optional communication.

ContextThe corrective action log is simultaneously a clinical quality document and a legal record. It must answer five specific questions: what test parameter failed, when the failure was first identified, who was notified and at what time, what service action was performed and by which technician or engineer, and what the re-test results showed. This document is reviewed at ACR accreditation site visits and may be independently requested during state radiation control program inspections or Joint Commission surveys. Cross-reference corrective action logs to the specific monthly QA packet they relate to — reviewers expect to find them together in the same binder or electronic folder.

ContextMost state regulations and all ACR programs require a supervising radiologist's signature on monthly QA records as evidence of physician oversight of the quality program. Email-only approvals are generally insufficient — a physical or electronic signature with a date-and-time stamp is the expected standard. If the responsible radiologist is unavailable on the day QA is run, document in the log the date that signature was subsequently obtained and by which physician. For multi-scanner facilities, each individual scanner requires its own separately signed record — a single physicist or radiologist signature applied across multiple units' records is routinely rejected as inadequate during ACR accreditation review.