Application of artificial intelligence in the analysis of asbestos fibers

Our plain-English summary of Application of artificial intelligence in the analysis of asbestos fibers, by R. Lee, D. Van Orden, S. Blanda, et al (Frontiers in Public Health, 2025).

The study

• Authors: R. Lee, D. Van Orden, S. Blanda, J. Mihalick, D. Bickford, P. Metsch
• Journal: Frontiers in Public Health
• Published: 2025;13:1584136
• Type of article: Review
• Cited by: 2 articles, as counted by NIH iCite
• Read it: PubMed · Publisher · Free full text

Abstract

Automated asbestos fiber detection and identification has been the goal of asbestos microscopists for decades. The advent of inexpensive memory, fast digital processing, machine learning, and microscope automation provide the enabling platform for success. This paper will review recent developments in fiber detection and identification by PCM and SEM and will present recent progress in employing artificial intelligence in the TEM classification of asbestos and non-asbestos amphiboles in the evaluation of elongated minerals in raw materials. To date, this project has been self-funded.

Copyright © 2025 Lee, Van Orden, Blanda, Mihalick, Bickford and Metsch.

In plain English

This paper partly reviewed earlier work on automated and artificial intelligence (AI) methods for finding and identifying asbestos fibers with phase contrast and scanning electron microscopy. It also described the authors' own early work on an AI system for transmission electron microscopy that sorts amphibole particles into asbestos and non-asbestos forms. A preliminary set of reference mineral samples, judged asbestos or non-asbestos from hand samples, included tremolite, riebeckite, actinolite, grunerite and anthophyllite. Some of the available training set data had been uploaded into the developing AI system. The authors used a decision tree method, with fiber length, fiber width and elemental composition as features. A test data set comprised approximately 2,500 fibers. In the example the authors reported, 84% of classifications agreed between the AI and an expert. In a simplified tree limited to two levels, an aspect ratio below 16.5 was the first decision. The authors concluded that the preliminary data showed an AI system was viable. They said a larger, more robust set of training samples was needed, along with an accepted procedure for validating results.

We wrote this plain-English summary from the full article with the help of AI, and a second AI model checked each sentence against the article. It is not the authors' wording; the abstract and the article are their own account.

What the study did and found

• Design: Partly a review of published automated and AI fiber-detection work (PCM and SEM), and partly a description of the authors' preliminary development of a decision-tree AI classifier for TEM discrimination of asbestos and non-asbestos amphiboles. The journal labels it Original Research.
• Who was studied: A preliminary set of hand-sized reference mineral samples (11 listed in Table 1, e.g. Jamestown tremolite asbestos, Shinness tremolite non-asbestos, Wittenoon riebeckite asbestos, Brockman riebeckite non-asbestos, Penge grunerite asbestos, West Greenland anthophyllite non-asbestos). The article says some available training set data were uploaded into the AI system but does not say which samples. The test data set comprised approximately 2,500 fibers. No dates are given.
• Exposure: No human exposure was studied. The materials were amphibole minerals in asbestos and non-asbestos habits; the preliminary reference list included tremolite, riebeckite, actinolite, grunerite and anthophyllite. Chrysotile was named among the regulated minerals, but the initial training sets focused on amphiboles. Particles were examined by TEM with elemental analysis (EDS) and electron diffraction (SAED). The decision tree features were fiber length, fiber width and EDS spectrum results.
• Main results:
  • Some of the available training set data, together with previously developed discriminant functions, were uploaded into the developing AI system.
  • In the example the authors reported, 84% of classifications agreed between the AI and an expert. The article does not give the number of particles behind this percentage.
  • The authors described their preliminary decision tree results as showing high classification accuracy.
  • One fiber from a training set, processed as an unknown, was classified as non-asbestos by the system. Two of the other discrimination procedures applied to the same fiber classified it as asbestos.
  • In a decision tree capped at two levels, an aspect ratio of less than 16.5 was the first decision. The authors interpreted aspect ratio as a key feature.
• The authors' conclusion: The authors concluded that the preliminary data showed the viability of an AI system, but also pointed to the need for a larger, more robust set of training samples.
• Limitations the authors state:
  • A larger, more robust set of training samples is needed.
  • Decision trees are prone to overfitting, especially on small datasets, and can be sensitive to small changes in the data.
  • The two-level tree shown for interpretability is not as accurate as a fully developed tree.
  • Where the expert and the AI disagreed, further investigation is needed to determine which characteristics caused the disagreement.
  • An accepted procedure to validate AI results is still needed.

Topics

Medical terms in this study, explained in our Lexicon: Amphibole Group.

Related studies

• Development of rapid and highly accurate method to measure concentration of fibers in atmosphere using artificial intelligence and scanning electron microscopy — Y. Iida, K. Watanabe, et al. (2021)
• Predicting Mesothelioma Using Artificial Intelligence: A Scoping Review of Common Models and Applications — M. Ram, M.R. Afrash, et al. (2025)
• Rapid fiber-detection technique by artificial intelligence in phase-contrast microscope images of simulated atmospheric samples — T. Yamamoto, K. Iwasaki, et al. (2024)
• Artificial intelligence-based diagnosis of asbestosis: analysis of a database with applicants for asbestosis state aid — K.B.W. Groot Lipman, C.J. de Gooijer, et al. (2023)
• Prospective validation of an artificial intelligence assessment in a cohort of applicants seeking financial compensation for asbestosis (PROSBEST) — I. Smesseim, K.B.W.G. Lipman, et al. (2025)
• Artificial Intelligence in Pleural Diseases: Current Applications and Next Steps — F. Karataş, Ö. Dikensoy (2026)

Browse more research in Frontiers in Public Health, or return to medical research.