7.1 Accuracy, Precision, and the Sources of Precision Error
Key Takeaways
- Accuracy describes closeness to the true value and is governed largely by scanner engineering and calibration, while precision describes reproducibility and is dominated by the technologist.
- A scanner with a systematic calibration offset can still monitor therapy reliably if its precision is good, which is why serial monitoring depends on precision rather than accuracy.
- Precision error sources divide into equipment characteristics, operator technique, and patient factors, with operator technique dominating in vivo.
- Scanner speed or acquisition mode affects photon statistics and therefore precision, so the same mode must be used at baseline and follow-up.
- Stationary phantom precision is typically better than 0.5% coefficient of variation, while in vivo precision including repositioning is commonly 1% to 2%.
7.1 Accuracy, Precision, and the Sources of Precision Error
Quick Answer: Accuracy is how close a measured BMD comes to the bone's true mineral content. Precision is how reproducibly the same measurement is obtained on the same subject. Diagnosis depends on accuracy; monitoring depends on precision. In clinical practice, precision error is dominated not by the scanner but by the technologist — through positioning, region-of-interest placement, and consistency of acquisition parameters.
The Distinction
The classic illustration is a target. Shots clustered tightly but away from the bullseye are precise but inaccurate. Shots scattered widely but centered on the bullseye are accurate but imprecise. Densitometry cares about both, but for different purposes.
| Accuracy | Precision | |
|---|---|---|
| Question answered | Is this the true value? | Will I get the same value again? |
| Error type | Systematic (bias) | Random (scatter) |
| Validated against | Ashed bone mineral weight, calibration standards | Repeated measurements of the same subject |
| Governed mainly by | Scanner engineering, calibration, beam hardening and soft-tissue algorithms | Technologist positioning and analysis; acquisition consistency |
| Clinical role | Diagnostic classification against reference databases | Serial monitoring and treatment decisions |
| Typical magnitude | Accuracy error of several percent is inherent to areal DXA | In vivo precision error commonly 1–2% |
Why Monitoring Needs Precision, Not Accuracy
Suppose a scanner systematically reports every BMD value 3% higher than truth. Every measurement on that machine is inaccurate by the same amount. But when you subtract a baseline from a follow-up, the constant offset cancels: the change is measured correctly. So long as the bias is stable and precision is good, the machine tracks therapy reliably.
Now reverse it. A perfectly accurate scanner with poor precision produces values that scatter by 4% around the truth. Every individual value is unbiased, yet you cannot distinguish a 3% therapeutic gain from noise. A stable bias is survivable; random scatter is not.
This asymmetry is why ISCD builds serial monitoring on a facility-specific, technologist-specific precision study and a least significant change derived from it, and why the diagnosis-versus-monitoring distinction recurs throughout the exam.
Accuracy: What Limits It
Areal DXA has inherent accuracy limits that no technique can remove:
- Projection. Areal BMD carries no depth information, so bone size influences the result.
- Soft tissue assumption. The composition of tissue overlying bone is estimated from adjacent baseline, and uneven fat distribution introduces error of a few percent.
- Beam hardening. The low-energy component of the spectrum attenuates preferentially with increasing patient thickness, shifting the effective energy.
- Reference database differences. Even a perfectly accurate BMD yields different T-scores against different databases.
Accuracy is maintained by the manufacturer's calibration, verified by internal calibration references and external phantom QC, and restored by service when QC detects drift.
Precision: The Three Named Sources
The ARRT outline organizes factors affecting accuracy and precision into equipment characteristics, operator characteristics, and patient characteristics.
1. Equipment Characteristics
- Scanner speed and acquisition mode. Faster modes deliver fewer photons per pixel, so quantum noise rises and precision degrades. Express or fast-array modes trade reproducibility for speed. Whatever mode is used at baseline must be used at follow-up — changing mode changes both noise and, on some systems, the calibration path.
- Scanner calibration and stability. Drift, shift, and mechanical wear degrade reproducibility. This is what daily QC exists to detect.
- Detector performance and mechanical repeatability of the scan arm.
- Software version. Analysis algorithm changes alter results; version consistency is part of precision control.
2. Operator Characteristics
This is the dominant source in vivo, and it is why precision studies are performed per technologist rather than per scanner.
- Positioning. Femoral neck internal rotation, femoral shaft alignment, spine centering and lordosis flattening, forearm alignment — each alters projected geometry and therefore area.
- Region of interest placement. Intervertebral marker placement, femoral neck box position and angle, global ROI boundaries. Manual adjustments made differently between visits create differences that are entirely analytic.
- Consistency of technique. Using a leg block one visit and not the next; choosing the right femur at baseline and the left at follow-up; excluding L2 once and including it later.
- Analysis independence during a precision study. Copying the prior analysis rather than re-analyzing artificially improves apparent precision and invalidates the study.
3. Patient Characteristics
- Body habitus. Very large and very small patients both degrade precision, through noise and soft-tissue baseline estimation respectively.
- Variant anatomy. Transitional vertebrae, short femoral necks, scoliosis.
- Pathology. Degenerative disease, calcification, and hardware move ROI boundaries unpredictably between visits.
- Low bone density. Reduced bone-to-soft-tissue contrast degrades edge detection, so precision is worse in exactly the osteoporotic patients being monitored.
- Ability to cooperate. Pain, tremor, and cognitive impairment increase motion.
Phantom Precision Versus In Vivo Precision
| Measurement | Typical precision | What it tests |
|---|---|---|
| Stationary phantom, no repositioning | CV typically < 0.5% | Scanner electronics and mechanical stability only |
| Phantom repositioned between scans | Slightly worse | Adds mechanical positioning variability |
| In vivo, patient repositioned | Commonly 1–2% | The complete clinical measurement — scanner, positioning, and analysis |
The gap between these figures is the technologist's contribution, and it is large. It is also the reason a facility cannot substitute phantom precision for an in vivo precision study when calculating least significant change: phantom precision describes the machine, while clinical decisions are made on patients.
Changes Affecting Scan Validity
The ARRT outline pairs "reproduce baseline study" with "changes affecting scan validity" — events occurring between two studies that break comparability even when the follow-up technique was flawless:
| Change | Effect on validity |
|---|---|
| Different scanner, manufacturer, or model | Values not interchangeable without cross-calibration |
| Different acquisition mode | Different photon statistics and, on some systems, a different calibration path |
| Software or reference-database update | Analysis algorithm or derived scores shift with no patient change |
| Different side scanned (hip or forearm) | Right and left differ systematically |
| Different vertebral exclusions or ROI placement | A different quantity is being measured |
| New fracture, hardware, cement, or prosthesis | The affected site or level is no longer evaluable |
| New or progressive degenerative change or vascular calcification | Adds mineral over time, mimicking a gain |
| Large weight or body-composition change | Alters the soft-tissue baseline estimate |
| A different forearm length entry | Percentage-defined regions land on different bone |
When any of these has occurred, the correct action is to state the limitation in the report rather than produce a number the reader will assume is comparable.
Reducing Precision Error
- Standardize positioning with written protocols, and use the same positioning aids every time.
- Reproduce the baseline exactly on follow-up: same scanner, same mode, same side, same ROI placement, same exclusions.
- Record deviations — an unusual position documented is reproducible; one that is not documented is not.
- Re-analyze rather than copy during precision studies.
- Perform daily QC and act on out-of-tolerance results before scanning patients.
- Retrain when precision fails the ISCD minimum standard, then repeat the precision study.
A scanner is found to report all BMD values approximately 3% higher than the true mineral content, but its precision is excellent. What is the practical consequence?
Why are ISCD precision studies performed for each individual technologist rather than once per scanner?
A follow-up spine scan is acquired in express mode because the department is running late; the baseline used standard array mode. What is the concern?