Machine Learning-enhanced background correction of fluorescence microscopy images

Dear Imaging Community,

As you certainly know the biggest issue when analyzing colocalization in fluorescence microscopy images is the reliability of results. This is because reliability depends on several technical factors, most importantly on the proper handling of omnipresent background noise in images, which is frequently very difficult to do.

After long development and careful testing, we are pleased to introduce CoLocalizer Pro 6 with Machine Learning-enhanced background correction.

We used tenths of thousands of real fluorescence microscopy images obtained using the three most common fluorescence microscope modalities (wide-field, two-photon, and confocal) to train a ML model to recognize different noise levels and then predict those levels on custom images. The released model achieves more than 90% prediction accuracy. It is integrated via Core ML, a blazingly fast new framework. We explain more about our ML model in the blog post:

To try ML Correction on your images, go to our webpage and download the latest version of CoLocalizer Pro for Mac app: Compare calculation results with and without ML Correction. You may be surprised how different they are.

Here is a quick demo of ML Correction in action:

If you have any questions, do not hesitate to ask.

Vadim Zinchuk
CoLocalization Research Software