Minimum detectable activity: an MDA calculation worked step by step
By Radiation Lab · Published · Reviewed
You counted a sample and no peak stood out. What is the smallest activity that count could reliably have seen? That figure is the minimum detectable activity, and this note works it through for Cs-137 on a gamma spectrometer, line by line, using the equations and the assumptions of the MDA calculator: Currie's detection limits with a paired blank. Then it reads three possible results against the limits, since that is where most confusion starts.
The measurement
The inputs below are illustrative but typical of a shielded detector and a one-hour count.
| Input | Value |
|---|---|
| Blank count rate in the 662 keV peak region | 2 cpm |
| Count time t, same for sample and blank | 3600 s |
| Full-energy peak efficiency ε | 3.0% |
| Emission probability Y, IAEA Livechart | 85.1% |
| Sample mass | 0.5 kg |
| Coverage k, for 5% false positives and 5% false negatives | 1.645 |
Step 1: the critical level LC
The expected blank count is B = 2 × 60 = 120 counts. A net count is the difference of two counts that each carry a variance of about B, which is where the 2 under the root comes from:
LC = k√(2B) = 1.645 × √240 = 1.645 × 15.49 = 25.5 counts.
This is the decision threshold, applied after the count. A sample with no Cs-137 in it would exceed it in only 5% of counts.
Step 2: the detection limit LD
LD = k² + 2LC = 2.706 + 2 × 25.5 = 53.7 counts.
That is the true net signal that would clear LC in 95% of counts. It describes the method, so it is worked out from the blank and needs no sample result. At k = 1.645 the expression is the familiar 2.71 + 4.65√B.
Step 3: from counts to becquerels
Each becquerel gives ε·Y·t counts: Y photons per decay, ε counts per photon, t seconds. Here that is 0.030 × 0.851 × 3600 = 91.9 counts per becquerel, so
MDA = LD / (ε·Y·t) = 53.7 / 91.9 = 0.584 Bq.
For the 0.5 kg sample that is 1.17 Bq/kg. To reproduce this on the calculator, note that a fixed count has one efficiency field, “Detector efficiency”, described as counts per emission from the source, and no field for emission probability or sample mass. Enter the product of ε and Y, 2.553%, as that efficiency, with 2 cpm, 3600 s and the 95% / 95% option, and it returns the same LC, LD and 0.584 Bq. Divide by the mass yourself.
Reading a result
LC is what a result is compared with. The MDA is not a reporting threshold, and treating it as one is a frequent error. Three possible net counts from this measurement:
| Net counts | Activity | Conclusion |
|---|---|---|
| 20 | 0.218 Bq | Not detected. Below L_C. |
| 40 | 0.435 Bq | Detected, though below L_D. |
| 70 | 0.762 Bq | Detected, and above L_D. |
The middle row surprises people. A net count of 40 is a detection even though its 0.435 Bq is below the 0.584 Bq MDA, because the MDA describes what the method reliably catches, not what it is allowed to report. In the first row, keep the measured value and its uncertainty alongside the words “not detected” instead of writing zero. MARLAP asks laboratories to report each result and its uncertainty as obtained, even a negative one, and never to report a result as “less than MDC”. That also keeps later averages honest.
Paired blank or well-known blank
The 2 in √(2B) assumes a paired blank: one blank counted for the same time as the sample, carrying as much counting noise as the sample does. Count the blank for longer and its share of the noise shrinks. With B still scaled to the sample's count time, MARLAP gives the variance of a net count on an empty sample as B(1 + tS/tB), and then LC = k√(B(1 + tS/tB)) and LD = k² + 2LC. A blank known so well that it adds nothing is Currie's “well-known” case. His table gives 3.29σB for it, and with the k² term kept that is about 2.71 + 3.29√B. A blank shorter than the sample pushes the other way.
| Blank | LC | LD | MDA |
|---|---|---|---|
| Shorter, tB = tS/4 | 40.29 | 83.3 | 0.906 Bq |
| Paired, tB = tS | 25.48 | 53.7 | 0.584 Bq |
| Longer, tB = 4tS | 20.15 | 43.0 | 0.468 Bq |
| Well-known | 18.02 | 38.7 | 0.422 Bq |
The MDA calculator implements only the paired row. If your blank is longer, its figure errs high and you can work the lower one from the expressions above. If your blank is shorter than the sample, its figure errs low, here by about 36%, and the unsafe direction is the one to watch. That row is only a guide, since its blank collects about 30 counts, below the range noted further down, where these expressions start to give too many false positives. Be careful about claiming the well-known row too. A background that drifts, or a blank from a different day, adds variation these Poisson expressions do not contain, and the lower figure is then not earned.
Where MDA figures go wrong
- Quoting LC as the limit. That gives 0.277 Bq here, too low by a factor of 2.1.
- Leaving out the emission probability. Y omitted, the answer reads 0.497 Bq, 15% low.
- Using the whole spectrum as background. Only counts in the peak region compete with the peak. At 200 cpm of total background the figure becomes 5.58 Bq, nearly ten times too high.
- Trusting an empty-shield blank for a real sample. A sample's own gamma lines, such as K-40 or the uranium and thorium series, can raise the continuum under the peak above the blank, and then the blank-based figure is too low for that count. Estimating the background from channels beside the peak also changes the 2 in √(2B).
- Expecting time to do what it cannot. Four hours instead of one gives 0.285 Bq, better by 2.05 times rather than four.
- Forgetting decay. The equations assume the activity holds steady during the count, and they refer to the time of counting. For I-131 counted 8 days after sampling, the MDA at the sampling date is 2.00 times larger. A count lasting a full half-life sees on average 72.1% of the starting activity. The MDA calculator applies neither correction. The decay calculator gives the first factor, which you divide by; the second is (1 − e−λt)/(λt), worked by hand.
- Very low blanks. MARLAP notes that the critical-level expression works well from about 100 blank counts, and can produce too many false positives below that. At B = 0 it gives LC = 0, so a single count would be called a detection.
- Not saying which convention was used. MARSSIM writes LD with 3 in place of k². With LC kept as above, that gives 54.0 counts, 0.55% higher; the validation report measures that difference. The form's 99% option, k = 2.326, gives 0.843 Bq. State the method beside the number.
- Treating the efficiency as exact. These equations, and the calculator, carry no uncertainty for ε or Y. MARLAP warns that the MDC comes out too low if the factor converting activity to counts is overestimated, so an optimistic efficiency flatters the result directly.
Check it yourself
- MDA calculator: L_C, L_D and MDA for a fixed count with a paired blank, and scan MDC.
- MDA validation report: MARSSIM's printed examples reproduced, and the difference in its constant.
- Radioactive decay calculator: the decay factor for taking an MDA back to the sampling date.
Sources
- Currie, L. A., Limits for Qualitative Detection and Quantitative Determination: Application to Radiochemistry, Analytical Chemistry 40, 586-593 (1968).
- Lindstrom, R. M., commentary on Currie (1968) with its Table I of paired and well-known blank expressions, in A Century of Excellence in Measurements, Standards, and Technology, NIST Special Publication 958 (2001), 164-166.
- Multi-Agency Radiation Survey and Site Investigation Manual (MARSSIM), NUREG-1575 Revision 1 / EPA 402-R-97-016 Rev. 1 / DOE/EH-0624 Rev. 1, August 2000. Chapter 6, Field Measurement Methods and Instrumentation.
- Multi-Agency Radiological Laboratory Analytical Protocols Manual (MARLAP), NUREG-1576 / EPA 402-B-04-001C, July 2004. Chapter 20, Detection and Quantification Capabilities: Equations 20.8 to 20.11 and 20.26, the blank-count range on page 20-16 and the reporting rule on page 20-12.
- IAEA Nuclear Data Section, Livechart of Nuclides — evaluated nuclear structure and decay data (ENSDF), International Atomic Energy Agency, Vienna.