A scanning and image processing system with integrated design for automated micronucleus scoring
Purpose: Our aim was to design a compact and cost-effective optical microscopic system for automated non-fluorescent micronucleus (MN) scoring whose performance can reach the accuracy of visual scoring with the help of minimal user interaction and also gives an option for fully automatic scoring with an accuracy suitable for triage purposes.
Materials and methods: The concept of Radometer MN-Series (RS-MN) microscopic system designed by Radosys was to develop hardware and software layers in parallel in order to optimize the performance in automated MN scoring. A MN assay slide is automatically scanned by the RS-MN then the binucleated cells and micronuclei are automatically identified. Processing 1000 cells takes 10–60 minutes by automatic scoring (scanning plus image processing) depending on sample quality and required accuracy. The manual revision of the cell gallery takes an extra 10 minutes per sample. Dose response curves are determined for manual, automatic and semi-automatic scoring methods.
Results: The combination of object-sensitive autofocusing method and the multi-layer image acquisition is able to reduce the minimum resolvable dose by 14%. The MN yields obtained from the manual, semi-automatic and automatic scoring methods are well correlating (Pearson’s correlation coefficients are between 0.977 and 0.998). In order to compare the reliability of the results of visual and automatic scoring, an extended analysis on uncertainty contributors was performed. For a dose of 1 Gy, the estimated relative uncertainty from the Poisson characteristics of MN yield is 17–19% for the manual and 20–38% for the automated scoring. Other uncertainty factors (differences in donor radiosensitivity, scorer performance, and sample preparations) can contribute to this error fall within a similar range: 3–16%. Taking into account all of the possible uncertainties, the minimum resolvable dose for the manual (0.48 Gy) is the two-thirds of that of the automatic scoring (0.61 Gy).
Conclusions: The results verify that the fully automatic mode of RS-MN is suitable for triage purposes. The performance of the user interacted semi-automatic mode is comparable with the reference manual scoring. Its performance reaches up to other non-fluorescent automatic systems and offers a compact and cost-effective alternative.
