Diagnostic Imaging Phantom QA Validation Project

This project builds a diagnostic imaging phantom quality-assurance validation package. The goal is to produce a reviewable engineering deliverable: intended imaging task, controlled acquisition setup, phantom measurements, image-quality calculations, dose-output checks where ionizing radiation is involved, uncertainty guard bands, trend review, failure modes, corrective actions, and release decision.

The project is educational and engineering-focused. It is not clinical advice, not regulatory advice, not a patient-specific imaging recommendation, and not a substitute for qualified medical physics, clinical engineering, radiology, institutional, or manufacturer procedure review.

Project Objective

Prepare a phantom QA validation package for a diagnostic imaging system after installation, software update, detector service, or recurring constancy review. The final deliverable should answer:

1. Which diagnostic or measurement task is being protected by the QA check?
2. Which acquisition protocol, reconstruction version, display path, phantom, and measurement method are controlled?
3. Do spatial sampling, contrast-to-noise ratio, uniformity, geometric accuracy, dose output, latency, and data integrity meet the acceptance criteria?
4. Are the results far enough from limits after measurement uncertainty is considered?
5. Does the trend show drift even when the current single measurement passes?
6. Which failure modes require corrective action before clinical or laboratory release?

The output is not only a table of phantom numbers. It is a release package that lets an engineering review board see what was measured, why it was measured, which task it protects, how uncertainty was handled, and what action follows from the evidence.

Project Boundary

Use this simplified boundary.

Item	Project value
System type	diagnostic imaging system with phantom QA workflow
Main example modality	CT-like cross-sectional imaging system
Transferable checks	ultrasound, optical, detector-array, and image-guidance systems after changing modality-specific limits
Intended engineering task	preserve task-based image quality after configuration change or recurring QA
Users	trained technologists, clinical engineers, imaging physicists, service engineers, and supervising clinicians
Evidence boundary	phantom acquisition, raw or reconstructed image record, measurement worksheet, dose-output evidence, version record, trend chart, deviation log, and release decision
Out of scope	clinical diagnosis, patient-specific dose estimate, regulatory submission, purchasing decision, and complete modality-specific standard procedure

The simplified numerical example uses a CT-like phantom because it allows clear calculations for spatial sampling, contrast-to-noise ratio, geometric scale, uniformity, and radiation output. The same structure applies to ultrasound or optical diagnostics when the metrics are replaced by depth accuracy, axial and lateral resolution, probe-element checks, optical power, detector saturation, wavelength calibration, or task-specific reference images.

Baseline Requirements

Use the following simplified requirements for the project.

Requirement	Acceptance criterion
Configuration control	scanner, protocol, reconstruction, workstation, display preset, phantom and analysis version recorded
Pixel size for the task phantom	at most 0.50\ \text{mm/pixel}
Target sampling	at least 4 pixels across a 2.0\ \text{mm} target
Low-contrast phantom CNR	at least 4.0 after uncertainty guard band
Image uniformity deviation	at most 8\% after uncertainty guard band
Geometric scale error	at most 1.0\ \text{mm} after uncertainty guard band
Radiation output difference, if applicable	at most 10\% from expected output after uncertainty guard band
Acquisition-to-review latency	less than 250\ \text{ms} for image-guided workflow
QA drift trigger	more than 10\% CNR decrease from commissioned baseline requires investigation
Data integrity	no missing image slices, corrupted metadata, or untraceable software version in the QA record

These are simplified engineering criteria. Real diagnostic imaging QA must use modality-specific standards, manufacturer procedures, local radiation protection rules, institutional acceptance limits, qualified reviewer judgement, patient-safety controls, and task-specific clinical evidence.

Deliverables

Prepare these deliverables before the release review:

1. QA protocol basis and intended task statement.
2. Configuration record with scanner, source, detector, protocol, reconstruction, display, phantom, software, and analysis versions.
3. Phantom acquisition record and image set.
4. Image-quality worksheet with pixel size, target sampling, CNR, uniformity, geometry, dose-output check where applicable, and latency check where applicable.
5. Uncertainty and guard-band table.
6. Trend chart against commissioned baseline.
7. Failure-mode and corrective-action matrix.
8. Release decision with accepted scope, restrictions, open actions, and next QA interval.

The deliverable should be reproducible. Another competent engineer should be able to rerun the same acquisition and understand why the decision was made.

Step 1: Lock the Configuration

Before measuring the phantom, lock the configuration.

Configuration item	Example record
Scanner identifier	CT-02
Tube-output or source mode	routine abdomen QA mode
Field of view	220\ \text{mm}
Image matrix	512 \times 512
Reconstruction kernel	standard soft-tissue QA kernel
Slice thickness	2.5\ \text{mm}
Phantom	task phantom ID P-CT-14
Analysis script	image-qa-v3.2
Workstation display preset	QA window/level preset
Software change under review	reconstruction patch R-2026.06

Engineering Comment

Configuration control is not paperwork decoration. Image quality can change with field of view, reconstruction kernel, slice thickness, detector calibration, display settings, analysis region, phantom position, and software version. If the configuration is not locked, a pass result may not correspond to the released clinical or laboratory state.

Step 2: Spatial Sampling Check

The phantom scan uses:

and:

Pixel size is:

Therefore:

The task target diameter is:

The number of pixels across the target is:

Acceptance Check

Quantity	Result	Criterion	Status
Pixel size	0.430\ \text{mm/pixel}	\leq 0.50\ \text{mm/pixel}	pass
Target sampling	4.65 pixels	\geq 4 pixels	pass

Engineering Comment

The target is sampled by more than four pixels, so the configuration passes the simplified sampling criterion. This does not prove true spatial resolution. Point-spread function, focal spot size, detector aperture, reconstruction kernel, motion, slice thickness, partial-volume effects, and segmentation method can still reduce effective resolution. The result is a configuration screen, not a universal resolution claim.

Step 3: Contrast-to-Noise Ratio

The phantom contains a low-contrast insert. The measured values are:

for the target region, and:

for the background region. Background noise is:

Use:

Difference in mean signal:

Contrast-to-noise ratio:

The expanded uncertainty of the CNR estimate from ROI placement, repeat acquisition, phantom reference, and analysis repeatability is:

For a lower-bound requirement, use the guarded value:

Thus:

Acceptance Check

Quantity	Result
Measured CNR	4.47
Expanded uncertainty	0.25
Guarded CNR	4.22
Acceptance limit	\geq 4.0
Decision	pass

Engineering Comment

The CNR passes even after the uncertainty guard band. This is stronger evidence than comparing the nominal value alone. If the measured CNR had been 4.10 with U_{CNR}=0.25, the nominal value would pass but the guarded value would fail. A phantom QA report should make that distinction visible.

The result still applies only to this phantom contrast, acquisition protocol, reconstruction version, display path, and analysis method. It should not be generalized to all diagnostic tasks.

Step 4: Uniformity Check

Use five regions of interest: center, north, south, east, and west. The mean image values are:

Region	Mean value
Center	1020
North	975
South	1046
East	966
West	1008

Define maximum uniformity deviation relative to the center:

The largest deviation is from the east region:

Therefore:

The expanded uncertainty in the uniformity deviation is:

For an upper-bound requirement, use:

Then:

Acceptance Check

Quantity	Result
Measured uniformity deviation	5.29\%
Expanded uncertainty	1.0\%
Guarded deviation	6.29\%
Acceptance limit	\leq 8.0\%
Decision	pass

Engineering Comment

The uniformity result passes with margin. The report should still include the phantom position, exposure mode, reconstruction version, ROI size, and whether the same display or raw-image basis was used as the commissioned baseline. A uniformity problem can come from detector calibration, beam hardening, source output, reconstruction correction, phantom centering, or analysis method.

Step 5: Geometric Accuracy

The phantom contains a calibrated distance:

The measured distance in the image is:

Geometric error is:

Therefore:

Use expanded uncertainty:

For a two-sided tolerance, compare:

with the allowed tolerance:

Guarded error:

Acceptance Check

Quantity	Result
Reference distance	100.0\ \text{mm}
Measured distance	100.6\ \text{mm}
Error	0.6\ \text{mm}
Guarded error	0.9\ \text{mm}
Acceptance limit	\leq 1.0\ \text{mm}
Decision	pass, but close to limit

Engineering Comment

The geometric check passes, but the guard-banded result is close to the limit. The release package should flag it for the next QA interval. If the imaging task includes quantitative sizing, implant fit, surgical guidance, radiotherapy planning, or serial measurement, this margin may be too weak even though it passes the simplified project criterion.

Step 6: Radiation Output Check

For ionizing imaging systems, image-quality release should not ignore output. The expected scanner-reported output for the QA protocol is:

The current output is:

Relative difference is:

Substitute:

Use expanded uncertainty:

For an upper-bound output-difference criterion:

Therefore:

Acceptance Check

Quantity	Result
Expected output	12.0\ \text{mGy}
Measured output	12.7\ \text{mGy}
Difference	5.83\%
Guarded difference	7.83\%
Acceptance limit	\leq 10\%
Decision	pass

Engineering Comment

The output is within the simplified tolerance after uncertainty is included. This check does not estimate patient dose and does not justify a clinical protocol by itself. It verifies that the QA acquisition output is not drifting away from the configured basis. Dose, image quality, task performance, and patient size must be controlled together.

Step 7: Latency and Data Integrity Check

For image-guided workflows, acquisition-to-review latency matters. Use this simplified timing budget:

Segment	Time
Detector readout	42\ \text{ms}
Reconstruction and correction	86\ \text{ms}
Network transfer	18\ \text{ms}
Workstation display update	31\ \text{ms}

Total latency:

Acceptance limit:

Acceptance Check

Quantity	Result
Total latency	177\ \text{ms}
Acceptance limit	250\ \text{ms}
Missing slices	none
Corrupted metadata	none
Version traceability	complete
Decision	pass

Engineering Comment

The latency result passes for the simplified image-guided workflow. If the system is used only for offline review, the latency criterion may be less important than archival integrity and workstation compatibility. If the system guides a procedure, latency, jitter, registration error, display update behavior, and user response become part of patient-safety evidence.

Step 8: CNR Trend Review

Single-test acceptance is not enough when drift matters. The commissioned baseline CNR was:

The current measured value is:

Percent decrease is:

Therefore:

The local investigation trigger is a decrease greater than:

Acceptance Check

Quantity	Result
Baseline CNR	4.8
Current CNR	4.47
Decrease	6.88\%
Investigation trigger	>10\%
Decision	no immediate drift investigation required

Engineering Comment

The current CNR passes both the absolute criterion and the drift screen. The trend should still be kept. A future value could remain above the absolute limit while showing sustained drift that points to detector gain change, source output change, reconstruction update, phantom positioning, environmental variation, or analysis-script change.

Step 9: Uncertainty Budget

Summarize the uncertainty sources used in the guard bands.

Measurand	Dominant sources	Expanded uncertainty used
Pixel size	field-of-view record, matrix record, reconstruction geometry	negligible for this simplified check
CNR	ROI placement, repeat acquisition, phantom insert tolerance, reconstruction noise texture, analyst repeatability	0.25
Uniformity deviation	ROI placement, phantom centering, detector correction repeatability, background drift	1.0\%
Geometric distance	phantom reference length, pixel interpolation, edge threshold, slice position	0.3\ \text{mm}
Radiation output difference	scanner display, dosimeter calibration or scanner output comparison, repeatability, setup	2.0\%
Latency	timestamp resolution, clock synchronization, display update detection	10\ \text{ms} if close to limit

Engineering Comment

The uncertainty budget does not need to be more complex than the decision requires, but it must cover the sources that can change the release decision. A small CNR margin or geometry margin needs a better uncertainty basis than a result far from the limit.

Step 10: Failure Modes and Controls

Connect the QA result to failure modes.

Failure mode	Engineering effect	Detection method	Control or response
Detector gain drift	CNR loss, nonuniformity, false texture	CNR trend, uniformity map, service calibration record	recalibrate detector, repeat QA, compare baseline
Reconstruction update changes noise texture	apparent CNR pass but task performance changes	version control, phantom comparison, user review	lock algorithm version, repeat task phantom, document release scope
Phantom mispositioning	false geometry, uniformity or CNR change	setup photo, alignment marks, scout record	repeat acquisition with controlled positioning
Dose output drift	exposure risk or image-quality shift	output comparison, dose dashboard, QA dosimetry	service check, radiation protection review, repeat output measurement
Display preset mismatch	user sees different contrast than QA evidence	display configuration record, workstation audit	lock display preset, verify DICOM or workstation settings
Metadata or slice loss	QA record no longer traceable	archive validation, checksum or image count check	reject record, repeat acquisition, investigate data path
EMI or network disturbance	corrupted transfer, delayed display, intermittent errors	event log, timing record, repeat under load	isolate source, repeat validation under representative load
Incomplete user workflow validation	phantom passes but clinical use remains risky	simulated-use check, handoff review	restrict release until workflow evidence is complete

Engineering Comment

The failure-mode table prevents a common QA mistake: treating a phantom pass as a universal system pass. A phantom verifies a controlled slice of performance. The release decision must still consider software, display, workflow, data integrity, maintenance state, and intended use.

Step 11: Release Matrix

Use a release matrix to make the decision explicit.

Evidence item	Result	Release meaning
Configuration record	complete	QA result is traceable to the released setup
Pixel size and target sampling	pass	acquisition geometry supports the simplified task screen
CNR with uncertainty guard band	pass	low-contrast metric has margin for the stated phantom task
Uniformity with uncertainty guard band	pass	detector and correction response are acceptable for the simplified check
Geometry with uncertainty guard band	pass, close to limit	release allowed, but next QA must monitor scale error
Radiation output check	pass	output agrees with the configured QA basis within tolerance
Latency and data integrity	pass	image-guided review path meets simplified timing and traceability criteria
CNR trend	pass	no immediate drift investigation required
Failure-mode review	open monitoring action for geometry margin	release with surveillance item

Final Decision

The system passes the simplified diagnostic imaging phantom QA validation package for the stated CT-like phantom task and controlled configuration. The release should be limited to the protocol, reconstruction version, display path, phantom method, and intended task recorded in this package.

The release package should include one surveillance action: review geometric scale error at the next QA interval because the guard-banded value is close to the acceptance limit. If the next result moves toward the limit, the team should check phantom positioning, calibration geometry, reconstruction scaling, table motion, edge-detection method, and service history before widening clinical use.

What the Final Report Should Contain

A complete report should contain:

1. intended task and acceptance criteria;
2. configuration record and version control;
3. phantom setup and acquisition evidence;
4. image-quality calculations with units and assumptions;
5. uncertainty budget and guard-band decisions;
6. trend comparison with baseline;
7. dose-output evidence where ionizing radiation is involved;
8. latency or workflow evidence where real-time review matters;
9. failure-mode and corrective-action table;
10. signed release decision, restrictions, open actions, and next QA date.

Common Mistakes

Common mistakes include:

• reporting CNR without phantom, ROI, reconstruction, and task context;
• treating pixel size as true resolution;
• passing a nominal value that fails after uncertainty is considered;
• ignoring trend drift because the current value still passes an absolute limit;
• changing the reconstruction algorithm without repeating task-based QA;
• validating a display path different from the released workstation path;
• recording dose output without connecting it to image quality and protocol control;
• omitting data integrity, metadata, software version, or archive checks;
• releasing a system for all use cases after testing only one phantom task.

Engineering Takeaway

Diagnostic imaging QA is a controlled evidence package, not a visual impression. A strong phantom validation project ties image-quality metrics to an intended task, protects the configuration from silent drift, applies uncertainty guard bands, watches trends, and states exactly what the release decision covers. The engineer’s responsibility is to make the boundary of the evidence explicit.