Image segmentation and paired shapes asymmetry quantification: An application in a Drosophila wing image se

The current process to identify wing pair shape asymmetry in Drosophila wing images contains multiple layers of potential measurement error. The image segmentation routine is a low-level method performed on a low resolution image set, and is prone to inaccurate edge detection in finding the wing's interior vascular structure and the exterior wing edge. An automated splining procedure on the segmentation result which yields the locations of several landmark points on the wing itself has several erroneous spline control points. The process to correct errors in the data requires both parameter tuning in the algorithm as well as manual correction of the segmentation and splining results. The in-production measures of asymmetry between Drosophila wing pairs are shown to be sensitive to these measurement errors. To reduce error in the segmentation step, several image segmentation methods are analyzed for use in developing a robust, efficient and automated segmentation algorithm for Drosophila wing image sets. Evaluation of the accuracy and efficiency of the methods is discussed, with a focus on the performance of multi-scale methods. A Frangi multi-scale segmentation is shown to more accurately locate the wing's interior vascular network. Additionally, an alternative principal components analysis of the variance structure in the image set is developed to isolate and quantify wing pair shape variation across the data set. This analysis replaces the splining process to identify locations of landmark points. Alternative measures of wing pair shape asymmetry are created from this analysis and an alternative measure of Directional Asymmetry (DA) is shown to reproduce existing benchmark measures of DA.