Automatizing the scoring of the cytokinesis-blocked micronucleus assay spares a lot of valuable time. The dose-effect relationship can be applied reliably for dose estimation, if the quality of the slides is the same from the perspective of the used image processing algorithm. This aspect brings in additional requirements against the quality of the slides compared to the conventional visual scoring. An add-in software was created to the non-fluorescent RS-MN automatic MN scoring system which is capable of measuring quantitatively the degree of typical anomalies. The image processing is less reliable when the presence of these anomalies is more frequent. The behavior of the designed sample quality parameters (SQPs) was tested on in vitro irradiated peripheral blood samples (0, 1 and 2 Gy) obtained from a healthy donor and also on samples from patients undergoing low dose-rate brachytherapy. We examined 20 different SQPs and identified two that are independent and correlate significantly with the error of the fully automatic MN frequency. One is related to the size of the cells and the other reflects the homogeneity of the environment. An equation was established which presents a connection between the error of the auto MN frequency and the SQPs. By adding a fourth cleaning step to the conventional sample preparation and changing the pre-dripping temperature of the slide, the SQP can be modified and consequently the sample quality can be improved. The gain in accuracy is 54 ± 10 MN per 1000 binucleated cells, which corresponds to the effects of 0.5 Gy. Around the lowest limit of detection (<0.5 Gy) it means a 50-100% drop in error of dose, which is significant. With sample quality harmonization, the positive predictive value was raised to 80 – 93% depending on the dose. With the technique described in this paper, the suitability for automated scoring of a micronucleus slide can be tested quantitatively and objectively. A method is presented with which in some cases the uncertainty of the assessed doses due to variance in sample quality can be decreased or if it is not possible its bias can be predicted. The proposed protocol leads to a more reliable estimation of dose. The SQPs are designed in a way that they have a potential to be adapted to similar systems.

Author