Figure 2

Overview of the automated steps to extract gene position information from images, which can then be employed for high-throughput studies. (a) The fluorescence image is acquired. (b) A multistage watershed segmentation algorithm creates a set of candidate segmentations. (c) A logistic regression assigns a probability to each candidate, screening out those with a low likelihood of being well segmented. (d) Examples of highly ranked vs. poorly ranked segmented objects. Blue: true segmentations of nuclei. Red border: automatically deduced candidate segmentations. Green/Red dots: FISH-labeled genes (e) Gene position measurements, such as the radial probability distributions, are made using the correctly segmented nuclei borders. (f) A confusion matrix illustrating potential outcomes of a binary classification. The red dotted line represents a candidate segmentation. False positives are the most critical source of error in a ranked retrieval of nuclei, potentially creating incorrect gene position measurements.