Computer Vision Algorithms for Quantifying the Growth and Behavior of Neurons Cultured on Nanofabricated Surfaces

Neuron growth on nanofabricated silicon surfaces, once understood, is fundamental to engineer circuits involving neurons and electronic components, leading to applications such as next-generation brain implants, and high-throughput neurologic drug screening assay systems. Precise quantification of neuronal growth and behavior from image samples requires computer vision algorithms for automatic tracing of neurons, and flexible algorithms for reliable registration and analysis of multi-fluorophore imagery. The tracing algorithms must be robust to high levels of clutter and common imaging artifacts, process discontinuities, and quantum noise, especially when live neurons are imaged. Finally, the large numbers of images that must be processed for a hypothesis test or assay call for computationally efficient algorithms. A fully automated neuron tracing algorithm based on the use of robust detection and estimation principles is described here that meets the above needs. It takes 2 seconds to trace and extract morphological statistics from a typical 1280 × 1024 image on a Pentium III, 1 GHz personal computer. It was validated against manually generated traces and the errors were in the range of 1-5% per image.