Automated neuron morphology reconstruction using fuzzy-logic detection and Bayesian tracing algorithms

Digital reconstruction of neuronal cell morphology from microscopy image data is an important task in many neuro-science studies. Since the quality of the images is typically low due to noise, imperfect staining, or uneven illumination, and the morphology of neurons can be very complex, reconstruction is often very challenging even for expert human annotators. Many (semi)automatic reconstruction methods have been proposed in recent years, but they are far from perfect, and the challenge remains to develop better methods. Here we introduce a new fully automatic neuron reconstruction method that combines fuzzy-logic based detection of critical points in the images and Bayesian probabilistic tracing between these points. The method was tested on 2D fluorescence microscopy images of real single neurons with corresponding manual annotations. Our method proves to be more accurate (smaller median error) and substantially more robust (smaller error variance) compared to an alternative state-of-the-art method based on all-path pruning.