Abstract
Recently, the asbestos health issues are becoming very large. We aim at the development of asbestos qualitative analysis support method in distribution staining method which used the phase contrast microscope. The automatic detection and counting system for particle and asbestos crystal are proposed in this paper. By using our proposed method, particle detection robust to change of the brightness and color of microscope picture and asbestos detection are realized. We perform actual counting experiment and compare our proposed method's result to inspector's conventional method to show the validity of proposed method.
Introduction
Asbestos remains a significant public health challenge, particularly during the demolition of older buildings. To ensure safety, technicians must perform a qualitative analysis (JIS A 1481) to identify asbestos crystals within building material samples. Currently, this is a manual bottleneck: an inspector must look through a phase-contrast microscope and count roughly 3,000 particles per sample. Because a human can only process about 10 samples a day, the process is a significant hurdle. While quantitative analysis has been automated, qualitative analysis is difficult because building materials contain a chaotic mix of various particles that look similar to asbestos under standard lighting.
Methodology
The researchers propose an automated pipeline leveraging the Dispersion Staining Method. In this technique, asbestos crystals exhibit a unique property: their color changes significantly when the polarizer of a microscope is rotated. Most other household or construction dust does not show this dramatic shift.
The proposed automated pipeline works as follows:
- Particle Extraction: The system first segments individual particles from the background. Since microscope lighting isn't always uniform, they use a robust background subtraction method to ensure they don't miss "faint" particles.
- Dual-Image Capture: The system takes two images of the same field of view—one before (Image A) and one after (Image B) rotating the polarizer.
- RGB Vector Analysis: For every extracted particle, the system calculates the average RGB values in both images. It treats these values as vectors in a 3D color space and calculates the angle between the vector from Image A and the vector from Image B.
- Classification: If the "angle" (the degree of color change) exceeds a certain threshold, the particle is classified as asbestos.
Experimental Results
The study evaluated the algorithm using 50 microscope images of Amosite (a common type of brown asbestos), from which 2,423 individual particles were extracted. The "ground truth" was established by a professional inspector.
Key results compared to the manual count of an expert inspector:
- Total Asbestos Crystals (Inspector): 136
- True Positives (System Correct): 110
- False Positives: 40
- False Negatives: 26
Discussion and Conclusion
Treating color as a 3D vector allows the system to be robust; even if a particle is relatively dark, a shift in its "hue direction" (angle) remains a strong indicator of asbestos. This proves that the dispersion staining color shift is a reliable enough feature to filter out the noise of similar-looking debris in building materials.
The primary limitation is the False Negative rate. Missing 26 out of 136 crystals is significant in a safety-critical field. Some misses occurred during the initial segmentation phase due to low contrast with the background. Future work will focus on improving initial segmentation, testing on other types of asbestos (like Chrysotile), and refining thresholding logic to reduce false positives, further streamlining the workflow for human inspectors.
Summary updated