The primary limiting factor behind the characterisation of the evolution in metal-alloy solidification has historically been the manual-intensive analysis process that has been required. Due to the low contrast, high noise and low dynamic range inherent to X-ray videos of in-situ solidification, traditional image segmentation approaches have proven to be ineffective. Exacerbating this issue is the X-ray nature of the images themselves, which do not just represent light but both material concentration and the position of any observed dendrites relative to the camera. Recently, an automated dendritic analysis system was developed, using both Machine Learning (ML) and bespoke, deterministic approaches within a modular architecture. The performance and speed of this system, from input-to-measurements, can be further improved via the introduction of supporting ML structures to existing, non-ML modules. These can assist existing algorithms by providing either corroborative or correctional results or, as training data allows, reduce the number of computationally demanding modules that need to be run. The results that will be presented demonstrate both the capacity to improve dendritic isolation via the use of a supporting ML as well the option of reducing overall processing times by forgoing the use of conventional system modules for certain dendrites, in favour of a pre-training Convolutional Neural Network (CNN).