Marc van Vijven, Application Scientist at CytoSMART - "During my Ph.D. research, I’ve harvested primary cells from clinical left-over material. Before being able to seed those cells at a pre-defined concentration into 3D constructs, I had to count the cells accurately. During this process, I wanted to be able to count cells quickly – minimizing the harm to the cells caused by maintaining them in suspension too long. Besides that, I wanted to increase the number of samples I could process simultaneously, enabling the screening of a large number of clinically relevant samples in a time-efficient way. Therefore, I chose to count the cells using a device rather than manually: I loaded the cells with the required reagents into a cassette, pressed the button, and obtained the cell concentration. However, after around six weeks, I noticed that many of my 3D constructs failed. Since one of my experimental read-outs was whether a construct failed or not, it took so long for me to suspect that this could be an experimental error rather than data. By checking the automated count a few times with an old-fashioned manual count, I found out that the automated count was a factor two to three too low, probably because the cells had the tendency to form clusters and the automated counter not being able to distinguish those. I hadn’t observed this before, due to the ‘black box' of the cell counting device: I had no idea how the device converted the cassette I inserted into the cell concentration. It cost me around 20 valuable clinical samples, and six Sundays (I had to stop my experiment every Sunday to have the equipment available for other lab members on Monday morning). The ideal solution for counting the cells would have been an automated counter providing me images it based its counts on and allowing me to apply corrections if I wouldn’t agree with the automated count. In that way, I would have had optimal control over my own experiment, combined with the speed and efficiency of automated counting. Now I could only blame myself for handing over control of such an important step in the protocol and depending too much on hidden steps inside a device."