Biliary tract autofluorescence cleaning for liver cryo-imaging data

Abstract

Cryo-imaging is an emerging biomedical imaging technique for studying cellular biodistribution in a mouse model. However, green autofluorescence, especially from bile ducts and the gall bladder, significantly interfered with green cell signals in fluorescent cryo-imaging data. This could make cell quantification in the liver data impossible. Recently, we observed that the autofluorescent signals tended to stay close to each other and formed dense clusters or structures in 3D space whereas the cells of interest were homogenously dispersed in the liver tissue. We propose that the autofluorescent signals could be rejected if they had a density measure in term of mean inter-particle distance (MIPD) above a threshold. We generated synthetic cell signals to test the algorithm. The cell signals were detected by applying the Mexican hat filtering and top-hat transformation to the fluorescent images; and followed by thresholding. The results of this process alone yielded detection precision and recall at 98% and 68%, respectively. With the density analysis, the detection results improved to 93% and 92% for precision and recall, respectively. This substantial improvement shows that the algorithm efficiently cleaned the false positives from autofluorescent signals. The cleaning algorithm worked great, especially on isolated cells in the liver data. In conclusion, we developed an algorithm for cleaning the autofluorescent signals, with minimal impact on cell signals for the first time. With the success, one should be able to analyze green fluorescently labeled cells in liver cryo-imaging data that was never possible before.