Analysing Complex Geological Materials with AZtecFeature & FeaturePhase

Geological structures are present on a variety of scales from kilometres to <1μm. These structures are produced by geological processes such as tectonic movement, volcanism and cooling. As a melt cools and solidifies, crystals sequentially exsolve from it into distinct mineral phases with different elements migrating to different phases, accommodated by differing crystal structures. This can cause a range of different characteristics within a rock, depending on the relationship between the minerals formed. There is a significant value in being able to measure, interrogate and understand these relationships as they provide a wealth of information on the rock in which they are found.

Geological samples can be challenging to analyse en-masse by automated approaches in the SEM. This can be for reasons related to the formation of the rock itself – such as mineral grains having very similar brightness, when imaged in the SEM, to neighbouring grains despite significant compositional differences. Here we discuss how automated analysis of challenging geological samples can be performed using a combined mapping and Feature analysis approach known as FeaturePhase.

Automated Feature analysis often uses backscattered electron (BSE) images for the identification and detection of grains from one another. This is because images of this type have a contrast that relates to the mean atomic number of the material under the beam – meaning that they can display compositional variation. Grey level thresholds are applied to these images to identify grain locations for subsequent analysis. This process can be hindered by various factors such as low BSE contrast between mineral phases and physical surface features such as cleavages, cracks and scratches. Such surface characteristics are clearly visible in BSE images and therefore have the potential to impact thresholding.

Example 1: A Single Grain

In the example below (Fig. 1A), two relatively bright grains can be seen to be next to each other in the centre of the image. The minerals forming the grains (Ilmenite (FeTiO₃) and Magnetite (Fe₃O₄)) have similar mean atomic numbers and therefore have similar grey levels in the BSE image. Many cracks that may have an effect on grey level thresholding are also present within both grains. Cracks often appear as darker than the host grain meaning that it may be treated as a grain boundary and multiple analyses may be performed instead of the real single grain being analysed. This is a common problem as it is often not desirable to make thresholds large enough to include cracks as darker thresholds are needed to detect other, real phases in the sample. However, the problem can be overcome by using EDS elemental mapping to detect the features by their composition instead of from the BSE image – effectively seeing through the cracks. In addition, the low contrast between the grains becomes irrelevant, allowing accurate analyses of the grains of interest to be performed. By utilising large area Ultim® Max detectors working at high count rates for this mapping, short live times per pixel can be used to ensure that throughput is maintained.

Fig. 1 - Electron images and EDS maps of magnetite and ilmenite. A. The different phases can be seen in the BSE image but grey level thresholding is made difficult by the low contrast between them and pervasive cracks. B-H: EDS maps of oxygen, silicon, titanium, aluminium, calcium, magnesium and iron. EDS mapping is not affected by the cracks or low image contrast in the same way as the BSE image

The acquired maps are then processed by our phase identification algorithm, AutoPhaseMap, which compares the individual element maps to determine what phases are present. Once identified, groups of neighbouring pixels of the same phase are combined to form features. These features are measured for morphology in the same way as "normal" features acquired on the basis of BSE image detection, and compositional information is extracted from the maps of the features. Classifications are applied to these phase-acquired features, as they are for "normal" features. An example of this is shown in Figure 2.

Fig. 2 - (Top Left) Automated EDS mapping, phase and feature extraction with AZtecFeature using FeaturePhase. The classification scheme (Top right) shows the number of features detected for each phase. (Bottom) Example of the combined morphology and compositional information recorded for the Ilmenite grain.

Example 2: Exsolution Lamellae in Pyroxenes

The analysis shown in case study 1 was a relatively simple case – working with a single field looking at one grain. In practice, we often require far more extensive analysis – looking at potentially thousands of grains over potentially hundreds of fields of view. The workflow that we just looked at can be extended to more complex scenarios. Here we consider exsolution lamellae of orthopyroxene (OPX) in clinopyroxene (CPX) as shown in Figs. 3 and 4. In this case, the OPX has exsolved from CPX (forming exsolution lamellae within the OPX grain) after their mutual crystallisation from an initial melt.

Whilst, in this region, scratches/surface damage are not a major issue, the grey levels of the different phases in the BSE image are very close, making FeaturePhase's mapping based approach particularly valuable. Fig. 3 shows an example of a single field BSE image and the associated EDS element maps. It is clear that although the BSE image gives some information, the EDS maps are far clearer in showing the distribution of the lamellae. In this case, the whole field of view has been mapped. However, it is also possible to perform a standard grey level thresholding of the sample and assign only certain thresholds to be mapped. This typically has the effect of greatly reducing the area to be mapped and therefore increasing throughput and is particularly useful where only certain phases within certain thresholds are hard to differentiate with the BSE image.

When a larger area consisting of multiple fields of view is mapped and montaged (Fig. 4), the extensive nature of the lamellae are shown and it is clear that large scale, automated analysis of this texture would give very useful information on the sample.

These maps were acquired automatically as part of a large area AZtecFeature run. Immediately after being acquired, they were processed with AZtec's FeaturePhase algorithm to identify phases and then recorded as features. This was repeated over the entire large area to form the image shown in Fig. 5. AZtecFeature's reconstruction algorithm was also applied to the dataset to reconstruct the larger features which were broken by field boundaries. With the grains all detected, it was then possible to interpret the data – an example of this is shown in Fig. 6 where the histogram shows the orientations of the long axes of the OPX grains – it is possible to see two dominant peaks – one for each of the major cleavage directions and the angle between them.

Fig. 3 - Single-field EDS map of fine OPX lamellae occurring in CPX. (A): The density variation is low, which is reflected on the BSE image. (B-D): EDS maps of oxygen, calcium and iron.

When a larger area consisting of multiple fields of view is mapped and montaged (Fig. 4), the extensive nature of the lamellae are shown and it is clear that large scale, automated analysis of this texture would give very useful information on the sample.

Fig. 4 - Multiple-field EDS map. This map is an extended area acquisition of the single field map of Fig. 3. A. The larger area covered shows additional phases are present and regions where surface damage is more of an issue

These maps were acquired automatically as part of a large area AZtecFeature run. Immediately after being acquired, they were processed with AZtec’s FeaturePhase algorithm to identify phases and then recorded as features. This was repeated over the entire large area to form the image shown in Fig. 5. AZtecFeature’s reconstruction algorithm was also applied to the dataset to reconstruct the larger features which were broken by field boundaries. With the grains all detected, it was then possible to interpret the data – an example of this is shown in Fig. 6 where the histogram shows the orientations of the long axes of the OPX grains – it is possible to see two dominant peaks – one for each of the major cleavage directions and the angle between them.

Fig. 5 - Large area AZtecFeature run with features coloured by class. The combination of automated feature analysis and phase analysis has produced a clear identification of both the host phase and the lamellae, as well as identifying additional minor phases

Conclusion

• The use of EDS mapping data for feature detection enables grains to be identified in geological samples where otherwise it would be difficult or impossible to do so.
• Once phases are identified, groups of pixels of the same phase are grouped to define Features. These are then treated in the same way as those acquired via the standard grey-level thresholding method.
• Sample preparation damage, cracks, topography etc. can effectively be "seen through" to get a true measurement of features of interest.
• The approach can be automatically applied to large areas and provides a wealth of information for understanding the rocks being studied.