I bet radiologists love you ( NOT!). Anyway, assuming you did find a radiologist willing and confident enough to turn over his job to a computer program, your #1 is not a problem. Not sure why you think it is. Who cares if the image are different sizes? It's normally not a problem.
For #2, you can try my interactive/visual thresholding application: http://www.mathworks.com/matlabcentral/fileexchange/29372-thresholding-an-image or else write your own routine. Most likely a simple intensity thresholding will not be robust to the wide variety of pathologies you can expect to encounter. But, what they heck, give it a try as a learning experience.
For #3, this will be dealt with as part of your thresholding algorithm in #2. It's not really a separate issue.
