New publication evaluates machine learning-automated analysis of μFTIR-images for Microplastics
In the 21st century, plastic, especially microplastic (MP) is omnipresent in nature, and its effect on human health and ecosystems remains widely unclear. Since the impact of microscopic plastic fragments depend on the exposure level and the material properties of these particles, it is essential to accurately evaluate the microplastic contamination with regard to polymer type, shape, and size.
Therefore, appropriate analytical tools and methods are needed since visual identification is extremely prone to bias with error rates up to 70%. In a joint project between scientists from the University of Bayreuth, the Vienna University of Technology and Purency, an easy to use analysis software was developed which reduces expert time, increases analysing speed and accuracy for making the overall process scalable. This is possible with the use of the imaging software Microplastics Finder (MPF) (www.purency.ai). In combination with spectroscopy, to be exact a focal plane array (FPA)-based micro-Fourier transform infrared (FTIR) imaging (www.bruker.com), the Microplastics Finder displays its strength due to its model-based machine learning approach. Here, the model is able to distinguish between more than 20 different polymer types (which covers 98% of currently produced plastics). Furthermore, it is also applicable to complex matrices, due to the large collection of spectra from a variety of sampling sites and matrix types (water, sediment, soil, compost, sewage sludge).
In this study, the performance and advantages of the used model under these demanding circumstances is presented based on eight different data sets. Once the analysis process is finished, the user may interactively assess and evaluate the list of detected MP in the particle editor which is also part of the software package. Therefore, the Microplastics Finder does not just increase the the quality of results but also provides an efficient workflow, enabling a high sample-throughput and making microplastics analysis scalable.
The full publication is open-access and can be read here.