Hi, I am a DL/ML expert, currently working as a deep learning specialist researcher at a mathematics research center, apart from freelancing work.
Currently, most of the best segmentation models are using the U-Net architecture, which is a Convolutional Neural Network (CNN) model. I have experience with implementing U-Nets in TensorFlow/Keras.
A relatively large set of labelled data is needed to perform learning, ideally from the same population then the actual images for inference. There are available datasets for research on the internet, I'm not sure whether there are too much that are available for a commercial application. I'll check if needed. Worst case manual labelling would be needed, which is time-consuming.
If needed, I can help deploying the application by packaging it into a Windows/Linux binary, even creating a basic GUI if needed. I can also deploy it on Amazon AWS servers, DigitalOcean servers, or your own servers.
I can create TFLite-compatible models, for deployment on mobile phones, embedded systems etc. if needed.
I have experience with the OpenCV library, using it together with ML libraries, in Python.