Weka Segmentation Errors

I have been working for a few weeks with trainable weka segmentation to count microglia on images. This is very standard mean os doing this for the field I am in. Normally, when I train images they do a very good job of identifying cells from background. However, recently I started training on larger stacks (with 6/7/8 images instead of under 4). With fewer images it is very accurate, but when I tried to do more images in my stack I got this very weird error, and despite all the retraining I tried, the weka couldn’t fix it. I have never seen anything like it. It is just not able to ID cells relative to background at all and speckles everything with no clear pattern (see first attached image) and am not sure how to work around. I also attached an image of what a good segmentation has looked like for me in the past (second attached image). Also attached example of original image (final attached image). Please help! Thank you so much!!