Tracing neurons through serial sections: Using knowledge of shape to improve performance.

This paper briefly presents TRACER-1 and illustrates an example of it succeeding for 6 sections and then failing. Then, the issue of incorporating more knowledge into the program is discussed, and the kind of knowledge needed for error recognition and recovery is explored for this one case of failure. An example of this kind of knowledge was incorporated into TRACER-2, and TRACER-2 successfully recognized and recovered from this error case. Other kinds of knowledge and issues related to this problem are discussed.