Abstract
BackgroundExtracting important descriptors and features from images of biological specimens is an ongoing challenge. Features are often defined using landmarks and semi-landmarks that are determined a priori based on criteria such as homology or some other measure of biological significance. An alternative, widely used strategy uses computational pattern recognition, in which features are acquired from the image de novo. Subsets of these features are then selected based on objective criteria. Computational pattern recognition has been extensively developed primarily for the classification of samples into groups, whereas landmark methods have been broadly applied to biological inference.ResultsTo compare these approaches and to provide a general community resource, we have constructed an image database of Drosophila melanogaster wings - individually identifiable and organized by sex, genotype and replicate imaging system - for the development and testing of measurement and classification tools for biological images. We have used this database to evaluate the relative performance of current classification strategies. Several supervised parametric and nonparametric machine learning algorithms were used on principal components extracted from geometric morphometric shape data (landmarks and semi-landmarks). For comparison, we also classified phenotypes based on de novo features extracted from wing images using several computer vision and pattern recognition methods as implemented in the Bioimage Classification and Annotation Tool (BioCAT).ConclusionsBecause we were able to thoroughly evaluate these strategies using the publicly available Drosophila wing database, we believe that this resource will facilitate the development and testing of new tools for the measurement and classification of complex biological phenotypes.Electronic supplementary materialThe online version of this article (doi:10.1186/s13742-015-0065-6) contains supplementary material, which is available to authorized users.
Highlights
Extracting important descriptors and features from images of biological specimens is an ongoing challenge
Information about the shape of the specimen is extracted from the configuration by removing variation in size, location and orientation of Sonnenschein et al GigaScience (2015) 4:25 the specimen, resulting in an explicit geometric representation of shape (Fig. 1) [5,6,7]
We describe the creation and implementation of a database of wing images from Drosophila melanogaster for the development and testing of such methods
Summary
Extracting important descriptors and features from images of biological specimens is an ongoing challenge. Widely used strategy uses computational pattern recognition, in which features are acquired from the image de novo. Subsets of these features are selected based on objective criteria. Geometric morphometrics and computational pattern recognition represent very different strategies for extracting and quantifying phenotypes from image data. Geometric morphometrics measures shape by using homologous landmarks (or curves) across specimens as features [3, 4]. These landmarks are determined a priori based on biological considerations of both homology and potential informativeness. Information about the shape of the specimen is extracted from the configuration by removing variation in size, location and orientation of Sonnenschein et al GigaScience (2015) 4:25 the specimen, resulting in an explicit geometric representation of shape (Fig. 1) [5,6,7]
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
