A novel algorithm for optimal image thresholding of biological data.

With the proliferation of both in vivo and in vitro microscopy techniques in the neurosciences, increased attention has been placed on the development of image analysis techniques. As experiments can produce large numbers of high bit depth images, automated processing methods have become necessary for handling these data sets. Thresholding, whereby a high bit depth image is converted into a binary image in order to identify a feature of interest, is one such standard automated technique; but the method of selecting an appropriate threshold value is far from standard. We present a novel algorithm, maximum correlation thresholding (MCT), that thresholds images accurately and efficiently without relying on any assumptions of the statistics of the image. As MCT produces thresholded images that preserve the most salient elements in the image, the algorithm performs as well as a trained user on a range of neurobiological data and in a variety of noisy conditions or when preprocessing steps preceded the thresholding operation. Our method will thus allow neuroscientists to automate image thresholding using a robust, computationally efficient algorithm, ultimately aiding in accurate image quantification and analysis.