Automated neuron tracing using the Marr-Hildreth zerocrossing technique.

Abstract

Semiautomatic systems designed to trace neurons in three dimensions are slow and require many hours of work on the part of a human operator to trace even a single neuron completely. Attempts at constructing fully automatic systems have met with only limited success because of the dual problems of slow speed of travel down a fiber and time/space limitations on the intelligence of the program guiding the decision making. A novel approach to automatic tracing has been designed and initial implementation experiments have been carried out. Consideration of the five theoretical issues of edge detection, decision making, accuracy, robustness, and speed has led to the use of the Marr- Hildreth preprocessing filter and to a simple fiber-following algorithm that directly utilizes the output of that filter. Experimental results are encouraging and suggest that the approach is worthy of continued research.