I would be more concerned about the photo recognition technology that is similar to what is used in the Amazon iphone app that allows the user to take a picture of something and then it searches amazon for something similar. From my experience, this works pretty well for common items. I can imagine someone creating a huge data base of normals for say every slice of a body CT (or organ) or head CT/MRI and then using the computer to compare each of the patient's slices to the normal database. The computer would then mark areas of abnormality and draw the radiologist's attention to these areas.
Eventually, you could have a database of abnormals specific to the area of where it finds the abnormality. For example- abnormality in the RLQ in area of appendix compares the patient's images to 1000 varieties/complications of appendicitis, cancer, backwash ileitis, etc... Abnormality in the pancreas- compares with database on pancreatitis, cancers, etc. Clinician suspects stone? The new renal protocol searches the area of the kidneys and ureters for a high density stone. Eventually, the computer could use key patient information (symptoms, age, vitals, labs) to narrow the differential and help focus its search. Of course, I think it would have a lot of problems for someone who has really messed up anatomy...i.e. whipple procedure or someone with mass trauma, but I think specific databases could be established for the different scenarios.
Even with something similar to this, I think there will be an intermediate phase where a radiologist is still absolutely essential. It is really hard to imagine a computer replacing us in our life time.