Robust classification of biological samples in atomic force microscopy images via multiple filtering cooperation

The morphological changes induced by engineered nanomaterials (ENMs) in cellular samples are a key clue to evaluate the impact of these materials on human health. The investigation of the complexity of the interaction among nanoparticles and cellular molecules requires cutting-edge instrumentation and dedicated procedures. To this regard, atomic force microscopy (AFM) is a leading imaging technique that has the peculiarities of high resolution and direct relationship with 3D cellular morphology. Expert human operators, however, are still required to manage most of the AFM-based analysis, thus introducing subjective bias and allowing a limited number of trials. The modality of interaction and the resulting topographic differences are often not a priori known. Also, the presence of artefacts in AFM images may affect the derived conclusions. In this work, we propose a robust and flexible strategy to analyse AFM topography images with single click-select actions by the human operator. The proposed system allows for not only morphological studies and quantification of the changes occurring in cellular samples in the presence of nanomaterials, but also for the investigation of diversified experiments in more flexible application domains. As a proof of concepts, samples of human EA. hy926 endothelial cells exposed to carbon nanotubes are used to demonstrate the effectiveness of the proposed solution. The system is also tested against various AFM-artefact and noise scenarios and the robustness of its discrimination capability is verified. (C) 2017 Elsevier B.V. All rights reserved.