With the advent of high power computers and high resolution video cameras, microscopy with image analysis is becoming one of the highest resolution techniques available today. Here we look at the development of a multimodal glass microsphere image analysis standard and evaluate the performance of one the latest automated instruments
AS MANY of the major manufacturers jostle for position as the ‘leading technology of the day’, end users can be confused as to the relative performance of the various instruments and with price tags in excess of £90,000, need some independent verification that they are selecting the right instrument for their application.
For this reason, there has been a significant interest in particle size calibration standards, in particular specific standards to exploit the high-resolution power of the technique.
The multimodal standard was prepared with eight individual and clearly defined peaks. Unlike conventional standards with Gaussian distributions, where only cumulative percentiles at 10%, 50% and 90% are reported, this standard not only covers percentiles from 5 – 95% but a value is assigned to each of the peaks, figure 1. Over 30,000 microspheres were analysed using a NIST traceable reference microscope. To ensure the peaks were assigned to the correct values, the standard was fractionated and the separated peaks analysed independently, Table 1.
Finally, a precision electroformed sieve analysis was performed to ensure the cumulative distribution overlaid that obtained from the microscope.
One of the disadvantages of using a simple laboratory microscope for particle size analysis is the time taken for a measurement, not only from the point of view of collecting and analysing the data but the mechanics of preparing and analysing the slides.
The 33,000 particles counted for certification took 8 hours to measure. Some recommend that over 50,000 particles should be counted, which makes the method impractical for routine measurement. However, 50,000 particles weigh about 75g so it is impossible to prepare a single slide of non-touching microspheres suitable for measurement.
|Figure 1: A Multimodal Test Certificate for Image Analysis|
To perform a certification, the bulk 200g sample had to be subdivided into weights below a gram using a spinning riffler. Data from the sub-sample were then combined until a statistically robust number of particles were counted. Another problem is the spherical nature of the particles. Unless they are constrained in some way, for example on transparent sticky tape, they simply roll off the microscope slide before they can be measured. A convenient way of measuring spherical particles over about 100 microns is to drop them from a vibratory feeder in front of the camera, figure 2.
The results from a HAVER CPA computer particle analyser using the complete 200g sample is shown in figure 3 and table 2. A total of 153,420 particles were counted so the statistics were very good. While the cumulative percentile values in table 2 are very close to the certified values, the real test of the accuracy and resolving power of an image analyser is whether the instrument can resolve and correctly assign the ‘finger print’ peaks in the multistandard.
|Table 1. Comparison of fractionated peaks with those from the complete 500 – 2000 micron multistandard|
Individual fractions (µm)
Sizes from certificate
The results in table 3 are exceptionally close for all the 8 peaks in the 500-2000µm multistandard and highlight the power of the image analysis technology.
The particle count necessary for optimum confidence in image analysis techniques is much debated. While 50,000 may be necessary to accurately define the extreme ends of the particle size distribution, it is not uncommon for some manufacturers of automated image analysis systems to insist that several million particles need to be counted.
This could take a considerable length of time, even for the most advanced systems, so the question must be asked ‘what is the optimum particle count?’
|Table 2. Comparison of HAVER CPA cumulative data* with the certificated values from a 500 – 2000µm image analysis standard (multistandard)|
| Certificate data
| HAVER CPA
|* 209.6g or 153,420 particles|
To see the bearing that particle count has on image analysis results, the nominal 200g sample was sequentially subdivided into two until the final weight was reduced to about 34g or approximately 23,000 particles. The graphical results in figure 2 show that the reduced weight/count has very little effect on the positioning of the 8 peaks, which are virtually superimposed. A summary of the effect of particle count and weight on the cumulative undersize analysis is shown in table 4.
As the particle count decreases below 40,000, the D90% and D50% remain relatively unchanged but the D10% values increase by about 4%, but whether this is a significant change is debatable as errors are bound to be introduced as a result of the subdivision process.
|Table 3. Performance of Haver CPA in analysing the individual peaks in a 500 – 2000µm image analysis standard (multistandard)|
| HAVER CPA
|* 209.6g or 153,420 particles|
Table 4 also clearly illustrates the speed of analysis that can be achieved with the latest automated image analysis instruments. The time required for the certification of the 500 – 2000mm multistandard of 8 hours was reduced to less than 4 minutes for the same particle count.
The availability of an independent and challenging particle size reference standard for Image Analysis now enables the prospective purchaser to unequivocally assess the performance the this new class of particle size analyser. High-speed electronics both in the computer and camera have reduced analysis times using conventional microscopy from several hours or even days down to a few minutes. The debate on the particle numbers then becomes largely irrelevant as 50,000 particles can be counted in only about 5 minutes.
|Table 4. Effect of particle count on image analysis results*|
|* HAVER CPA Image Analyser|
One of the biggest advantages of image analysis is in the ability to identify and size closely lying peaks in a sample and, provided in excess of 50,000 particles are counted, fine detail at either end of the distribution can also be determined.
Figure 2: The effect of particle count on peak position (HAVER CPA)
The free fall or dynamic class of image analyser as described above represents one of the highest speed methods of particles size analysis. One aspect that has not been dealt with here is the ability to determine particle shape as well as size. The subject of particle shape and the debate between dynamic and static analysis will be the subject of a future review.