Latest articles

Exposing the Silent Killer: New AI-based Method for Detecting Toxic Airborne Asbestos FibersAsbestos—a naturally occurring mineral made up of thin, microscopic fibers—is a known cancer-causing substance. Despite being banned in several countries around the world, asbestos is found in old buildings, and low levels of airborne asbestos can be detected in general environments. However, the concentration of asbestos fibers is reported to be much higher around demolition and construction sites, putting workers at risk of asbestos exposure and necessitating regular on-site inspections. Although traditional techniques such as “phase contrast microscopy” (PCM) and “scanning electron microscopy” (SEM) allow the detection of asbestos levels, they are often time-consuming and require manual effort. Therefore, their utility in on-site inspections is rather limited.   In an attempt to make asbestos fiber detection more rapid and efficient, a group of researchers from Japan combined SEM with an artificial intelligence (AI) system that allows the detection of thin fibers that cannot be seen on conventional PCM. They tested the usefulness of this hybrid system—which they called AI-SEM—on a sample of white asbestos. First, they obtained 108 images of the sample using SEM at a 10,000-times magnification and then trained their AI with 25 of these images. Then, the number of fibers in all 108 images were counted separately by a trained expert and the AI system.They found that the AI-SEM system could detect almost 88% of asbestos fibers, including those as thin as 0.06 micrometers. While this detection rate was similar to that observed after manual counting by a skilled analyst, the AI-SEM system achieved a 50-times faster processing speed, significantly reducing the time required for analysis.Hence, the findings show that AI-SEM has the ability to count thin fibers more efficientlty and accurately than conventional methods such as PCM and SEM. It could someday be used for measuring airborne fiber levels around construction sites and sites of accidents, thereby helping reduce exposure and ill-effects on human health. However, these findings are preliminary, and more research will be required to improve the performance and reliability of this system.