What do we mean by "automatic" ?

 A brief outline of the expectations and realities of "automated" systems, and the exciting future ahead of us.

Since its beginning in 1979, our company has supplied equipment to various scientific disciplines which avoids tedium and increases productivity, by automating certain complex processes.

In the case of fission track dating, this started with the automatic positioning of the sample under the microscope. This involved the precise location of points on the mica external detector which corresponded to specific locations on the grain mount. Counting was carried out manually in the early days, with the operator sitting at the microscope and observing the number of tracks in a given grid cell on the eyepiece graticule, and counting the number of tracks, usually with the aid of a hand-held "clicker" counter.

Later, we supplied enhanced equipment which incorporated a drawing tube attachment and a digitiser tablet. With this arrangement, the operator was able to use the digitiser mouse ("puck") to both count the number of tracks and measure their lengths.

With improvements in technology, and in particular in graphics cards, monitors, CCD cameras and ever faster computers, it became possible to carry out the counting and length measurements on-screen at the computer. This avoided the fatigue and headaches associated with long hours of looking into the microscope eyepieces. The addition of colour and the ever-higher resolution of digital cameras have made this process easier than ever.

What we have always been at pains to point out is that our equipment does not make the final decisions which a human expert must make about the entities being observed. Thus, in our radiation-protection application for instance, we do not decide what is an alpha track in CR39 and what is not. Neither, in FTD, do we make the final decision as to what is a fission track and what is not. That is a decision that should always be left to the expert.

What we do do to support the work of such experts is to simplify the decision-making process: first of all, by eliminating the fatigue associated with tedious and repetitive processes, and secondly, by presenting the operator with a reduced data set with a much higher likelihood of containing the desired elements than the initial data set of an infinite number of possibilities. We also document the recognised entities in a data table for later retrieval or re-examination. In our latest version, we in fact store a complete three-dimensional image stack, which is a record of the grain prior to destruction by the latest analytical tools (such as LA-ICP-MS). This allows the operator to return to the source data in the event of any later uncertainties or the need for further investigation or validation.

Thus our latest offering, the "automatic fission track counting" systems, certainly do offer full automation for the capture and counting of fission tracks, and the storage of the data tables and image sets. During that process, no operator intervention is required. However, there is obviously an initial setting-up process involved for each sample, as well as a final review and editing process. Computers have simply not yet evolved to the point where they can operate like a human brain – anyone who is involved with the field of Artificial Intelligence is well aware of this. So for instance, both in fission track dating and radiation protection, there exist instances of multiple overlaid tracks. In many cases, even a human operator can have difficulties in deciding how many tracks are involved, and often such track clusters are simply ignored. Our software makes an estimate of this number based on the measured track parameters, but always allows the operator to override such decisions.

In evaluating the overall value of a system such as ours, in comparison with manual execution of the processes involved, it should be kept in mind that just the process of object recognition and counting saves the operator a huge amount of time, and is also possible for densities well in excess of those a human operator is able to conveniently count under normal observation conditions. It is then up to the operator to eliminate the few false positives and false negatives which may result from this process. However, the volume of data involved at this point is considerably smaller than the huge volume of initial candidates.

A further attraction of process automation is the consistency of the machine. A human operator will often arrive at different counts for the same sample, depending on the level of interest, fatigue and physical wellbeing (such as visual acuity, etc.) By contrast, a machine may not arrive at a number which is exact (and neither will the human operator – the exact value of a large number of objects is often unknowable), but under identical circumstances it will always be consistent. That is an important plus.

The results obtained to date (and we do already have a number of these installations in place and being used) have shown that the results of the automatic count are in very good accord with those obtained by a human operator. Several people have commented along the lines that “this approach would never work for my samples”, to which the answer is that this system will work for any sample that a human operator can count, because it is always the human operator that is making the final judgement.  The differences from one sample to the next will be in the degree to which the operator is required to do this during the final review.  With very straightforward samples the operator will need to do very little, if any, editing of the automatic counting results, and in others relatively more.  In many samples the software produces excellent results with no operator editing whatsoever.