logo white

7.3 Emerging Technology for Asbestos Fiber Counting

Jordan Gruber

eToolkit Menu

Overview

 

Recent advances in robotics technology have allowed for the creation of new asbestos fiber counting products and methodologies that improve upon traditional manual analysis. The primary benefits of these new technologies are:​

  • Reducing risk to workers and the public by more quickly identifying areas containing dangerous concentrations of airborne asbestos.
  • Reducing project delays and risk of cost blowouts on commercial, residential, and government sites by providing faster turnaround of results.
  • Providing a more effective chain of custody for each sample analysis through automatic data collection and retention, thereby reducing risk of reporting errors or foul play.

​​“chain of custody”: Chain of custody (CoC), in legal contexts, is the chronological documentation or paper trail that records the sequence of custody, control, transfer, analysis, and disposition of materials, including physical or electronic evidence (Wikipedia).

In this chapter we will explore the MARVIN system for automated asbestos fiber counting, describe how it works, how it can be deployed in practice.

 

MARVIN system for automated asbestos fiber counting

Technology Description

​MARVIN is a product that has been developed by Frontier Microscopy over several years. It combines affordable microscope robotics hardware with artificial intelligence software to automated air filter analysis. Through automation MARVIN not only meets established standards for asbestos fiber counting but can also be configured to exceed requirements of the standards to provide more accurate and consistent results. Additionally, through clever software MARVIN can be operated by any non-scientist with just a few hours of basic training. The MARVIN system analyses common whole, or half air filter samples prepared onto microscope slides in any diameter up to 25mm.

 

MARVIN system

MARVIN Hardware

 

MARVIN Hardware
MARVIN Hardware
MARVIN Hardware
MARVIN Hardware

MARVIN Software

​Analysis of prepared air filter samples by MARVIN is broken into 5 main steps that will be discussed in more detail.

  1. System Calibration
  2. Macro Imaging Assessment
  3. Sample Digitization
  4. Digitized Sample Analysis

System Calibration

​
To ensure MARVIN can reliably perform sample digitization it must be calibrated at least once daily. System Calibration is easy to perform with the MARVIN interface guiding the operator through each step of the process. Through this process we ensure MARVIN can move to points of interest on the sample with high accuracy, and that phase contrast optics are correctly set so that asbestos fibers can be captured in images during sample digitization. To ensure high quality of analysis MARVIN will not permit the operator to perform an analysis unless calibration has been performed properly. With practice System Calibration can be performed in under a minute.

During calibration MARVIN checks phase contrast optics are correctly set by use of an installed Detection Limit Test Slide. This slide is specially produced to have very thin line etchings filled with resin that act as an optical analogue for asbestos fibers. Under normal brightfield microscopy these etchings cannot be seen but by the mechanism of phase contrast microscopy the etchings can be observed so long as the phase contrast optics have been correctly calibrated. An example image captured by MARVIN showing these etchings is presented below. For quality assurance and traceability MARVIN saves images of these etchings with each sample analysis.

For more information on phase contrast calibration refer to [ref].

 

phase contrast calibration

Macro Imaging Assessment

​Macro Imaging Assessment is a critical step of MARVIN analysis process that enables a non-skilled scientist to operate MARVIN. During this step MARVIN moves the prepared sample under a secondary Macro Camera which captures a 1x image of the sample for processing. From this image the MARVIN AI maps features of the sample such as the filter area, grid lines on the filter, and large dirt deposits or smears. From this map MARVIN applies quality criteria to either accept or reject the sample which would otherwise need to be performed by a skilled scientist. Lastly MARVIN generates a scan path of points to visit on the sample consisting of two types of points:

  1. Gridline intersection points – By most regional standards gridded air filters are mandated as the gridlines provide reference points for analysts to ensure they are observing particles on the filter, rather than on top of the coverslip, or on the back of the filter.
  2. Scan Points – MARVIN selects 100-500x points on the filter that will be visited during a scan and images captured for fibre counting analysis.

Without macro-imaging MARVIN cannot identify the position of grid line intersection points and therefore would be unable to ensure that images captured at scan points are actually on the filter and not on the coverslip. An example output of the Macro Imaging Assessment is shown below with gridline intersection points identified by pink crosses, scan points by green dots, gridlines by green, and valid filter area by blue.

 

output of the Macro Imaging Assessment

Sample Digitization

​Gridline Intersection Point Capture

​Gridlines exist on air filter samples to provide reference points for analysts to be able to ensure that they are looking at the correct depth where the air filter surface is, rather than the bottom of the filter or the top of the coverslip on the sample. Similar to a human analyst MARVIN uses these gridlines to ensure quality of sample scanning. Before collecting images for bire counting MARVIN moves to several points on gridlines across the sample and automatically focusses itself on them. At each point on the gridline the focus position is recorded to be used later during the point scanning for analysis. An example of a gridline is 400x total magnification is shown below.

 

Gridline Intersection Point Capture

Point Scanning for Analysis

​Using the previously attained positions of gridlines MARVIN performs point scanning where it will visit at least 100 points on the sample and capture images across the depth where particles on the air filter membrane are expected to be found. The position of each point visited is recorded and up to 10 images captured at each point but at varying depth. Unlike traditional analysis where images are not captured, and the physical sample is often discarded within weeks these images can be stored indefinitely. In addition, MARVIN can be configured to analyze far more than 100 points on the sample which only represents <1% of the total sample area. With more points captured MARVIN vastly increasing accuracy.

Digitized Sample Analysis

​MARVIN strictly applies the morphology counting rules as set out in the global standard.

MARVIN’s AI is very complex with a number of specialized algorithms that can accurately count for a variety of field samples. MARVIN can:

  • accurately separate particles from their glowing artefacts
  • separate overlapping respirable fibres
  • count accurately for field samples with medium-high particulate coverage.
Digitized Sample Analysis

 

MARVIN also provides more consistent sample results than humans can as it can apply some counting rules that are otherwise objective for humans. One such example is the ⅛ graticule coverage rule whereby a point should not be analyzed if more than 12.5% of the graticule is covered in particles. The reason for this rule is that if there are too many particles then otherwise countable respirable fibers may not be counted, thus resulting in a pass result when the sample should really be failed.

 

MARVIN consistent sample results

Results for analysis by MARVIN are output in fibers per valid fields visited. Industry generated results for the MARVIN AI can be seen below. This graph shows performance of the best MARVIN AI candidates over several weeks.

The manual data was produced by using our own software tool to count fibers in point image sets. Each MARVIN AI candidate analyzed these same point image sets to produce the overall count. For some samples such as 19-32 the trend is not linear which is expected as each sample varies in particle types present, particulate density, and background textures.

 

Improvement in MARVIN vs. Manual Analysis Results (All Weeks)

Benefits for Stakeholders

​The MARVIN system aims to become best practice for air filter analysis. Through automation MARVIN can be operated by almost anyone with little training required. The ma

​Regulatory & Public Health

  • Superior data retention. Digitized samples can be retained indefinitely and reviewed if required.
  • Transparency. Sample and analysis data may be accessible to regulators to ensure compliance with the approved analysis method.
  • Improve best practice. All accepted methods by human analysis extrapolate the total fiber count from <1% of the total sample area which can result in widely varying results between analyses. MARVIN can be configured to analyze a far greater area of the sample and reduce the uncertainty of the result.
  • Contamination events may be automatically logged with regulators so rapid action can be taken or enforced.
  • Reduce risk of human error due to fatigue or analyst training by the traditional method.

Commercial​

  • Rapid turnaround of results allows job sites to be cleared much more quickly than traditional offsite laboratory analysis.
  • Reduce risk of human error due to fatigue or analyst training by the traditional method.

References

  1. International Standard Organization (1986). Determination of the Number Concentration of Airborne Inorganic Fibers by Phase Contrast Optical Microscopy – Membrane Filter Method, 3rd Revision, ISO/TC, 146/SC2 WG5.

Sign up to our newsletter

Sign up
Skip to content