Graph cuts and neural networks for segmentation and porosity quantification in Synchrotron Radiation X-ray μCT of an igneous rock sample

Anderson Alvarenga de Moura Meneses; Dayara Bastos Palheta; Christiano Jorge Gomes Pinheiro; Regina Cely Rodrigues Barroso

doi:10.1016/j.apradiso.2017.12.019

Graph cuts and neural networks for segmentation and porosity quantification in Synchrotron Radiation X-ray μCT of an igneous rock sample

Appl Radiat Isot. 2018 Mar:133:121-132. doi: 10.1016/j.apradiso.2017.12.019. Epub 2017 Dec 21.

Authors

Anderson Alvarenga de Moura Meneses¹, Dayara Bastos Palheta², Christiano Jorge Gomes Pinheiro³, Regina Cely Rodrigues Barroso⁴

Affiliations

¹ Federal University of Western Pará, Institute of Geosciences and Engineering, Laboratory of Computational Intelligence, R. Vera Paz, s/n, Salé, CEP 68.035-110 Santarém, PA, Brazil; Postgraduate Program in Amazon Natural Resources, Federal University of Western Pará, Brazil. Electronic address: anderson.meneses@pq.cnpq.br.
² Federal University of Western Pará, Institute of Geosciences and Engineering, Laboratory of Computational Intelligence, R. Vera Paz, s/n, Salé, CEP 68.035-110 Santarém, PA, Brazil; Postgraduate Program in Amazon Natural Resources, Federal University of Western Pará, Brazil.
³ Federal University of Espírito Santo, Brazil.
⁴ Rio de Janeiro State University, Brazil.

PMID: 29367125
DOI: 10.1016/j.apradiso.2017.12.019

Abstract

X-ray Synchrotron Radiation Micro-Computed Tomography (SR-µCT) allows a better visualization in three dimensions with a higher spatial resolution, contributing for the discovery of aspects that could not be observable through conventional radiography. The automatic segmentation of SR-µCT scans is highly valuable due to its innumerous applications in geological sciences, especially for morphology, typology, and characterization of rocks. For a great number of µCT scan slices, a manual process of segmentation would be impractical, either for the time expended and for the accuracy of results. Aiming the automatic segmentation of SR-µCT geological sample images, we applied and compared Energy Minimization via Graph Cuts (GC) algorithms and Artificial Neural Networks (ANNs), as well as the well-known K-means and Fuzzy C-Means algorithms. The Dice Similarity Coefficient (DSC), Sensitivity and Precision were the metrics used for comparison. Kruskal-Wallis and Dunn's tests were applied and the best methods were the GC algorithms and ANNs (with Levenberg-Marquardt and Bayesian Regularization). For those algorithms, an approximate Dice Similarity Coefficient of 95% was achieved. Our results confirm the possibility of usage of those algorithms for segmentation and posterior quantification of porosity of an igneous rock sample SR-µCT scan.

Keywords: Geological samples; Image segmentation; Micro-Computed Tomography; Synchrotron Radiation; X-Ray.