Strange Outlines Detected in Images

Hi Guys,

I’m currently trying to analyse about 3,000 images with a considerable variation in cell coverage - from near confluent to only a few cells. Getting the thresholding right for the full dataset is proving a challenge, but I have a reasonably good pipeline up and running.Full Data Aquisition 02.cp (8.53 KB)
As I get good detection of my nuclei, I’m using an adaptive thresholding method. My substrates are not perfectly flat, so there is unfortunately a bit of focus drift towards the edge of each image. (10x objective, 900 micron across FOV). Kapur Adaptive seems to cope with this the best.

One problem I’m seeing, however, is square outlines being drawn where there are no objects? I wonder if this is a problem with my image files, and some sort of underlying hysteresis I can’t see.

The files are captured as 16bit TIFF files using ImageProPlus on an Olympus scope, QCam camera. As CPA wasn’t a fan of the 16bit tiffs, I converted them to 12bit using imagemagick.

My ultimate goal is to train the CPA classifier to detect and sort two cell types in co-culture. As the cell size is a key difference between my two cell types, I’m worried that these intermittent square regions and also the ‘sphagetti’ type edges might skew the object measurements and make it difficult.

Any help is much appreciated,

Hi Paul,

The adaptive method uses a square NxN pixel neighborhood to evaluate the threshold which can lead to artifacts in neighboring low-intensity regions; this issue was noted and a fix proposed here. However, this modification has not yet been released. A workaround for this problem is to determine a lower bound for the thresholding method such that discrepancy between neighboring dim regions is eliminated.


Hi Mark,

Thanks for the link, I’ll try playing around with the modifications suggested there and post some comparisons if worthwhile. Tweaking the lower threshold also improved things for now