Microplastic detection and characterization in different matrixes by computed tomography

Abstract Microplastic (MP) pollution represents a growing environmental threat, particularly in terrestrial and coastal ecosystems. Characterizing MPs is challenging due to their heterogeneous shapes, sizes, and chemical compositions, as well as their interactions with complex environmental matrices. This study evaluates X-ray microtomography (microCT) combined with linear regression models (LRM), a machine learning approach within artificial intelligence (AI), for high-resolution detection and morphometric characterization of MPs in soils. Controlled experiments used sandy, humus-rich, and mangrove soils spiked with polyethylene terephthalate (PET) fragments (~ 1 mm). MicroCT provided non-destructive three-dimensional imaging, from which volume, surface area, elongation, flatness, anisotropy, and sphericity were extracted. Image processing, including filtering and morphological corrections, enhanced particle delineation and data accuracy. LRMs predicted interdependent morphometric parameters, with performance assessed using mean squared error (MSE), mean absolute error (MAE), and the coefficient of determination (R 2 ). Results showed high predictive accuracy for sandy and humus-rich soils (average R 2 ≈ 0.93–0.94), whereas mangrove soils exhibited lower and less stable performance (average R 2 ≈ 0.69), reflecting matrix heterogeneity. These findings indicate that soil properties strongly influence microCT-based MP characterization and that linear regression may be unreliable in complex matrices; however, combining high-resolution imaging with predictive modeling provides a more robust approach for accurately assessing MPs across diverse environments, supporting improved environmental monitoring and informing methodological strategies tailored to specific substrates.