P. N. ANGEL. Automating the Identification of Species of Parasitic Wasp using Image Processing and Neural Networks.
Automatically locating taxonomic landmark points from image data is made difficult given the presence of natural deformation, specimen damage and unwanted debris. The problem is compounded when texture is present, which makes finding landmark points located on the boundaries of natural structures more difficult. To address this problem, a novel texture filtering technique is used to enhance those boundary features which contain salient landmark points. Once the texture has been removed, statistical models of the boundary structures can be created and these can then be used to locate landmark points in the original image. Once the landmark points have been found, the taxonomic features derived from them are applied to a back propagation neural network classifier for species identification. This process has been applied to the problem of automatically identifying species of parasitic wasp of the order Hymenoptera using SEM images of the anterior view of their heads. Five closely related species have been chosen, making it difficult to differentiate between them. The results have shown that a 95% landmark localisation accuracy and a 91% correct identification rate can be achieved using the proposed technique.
Neural Computing Research Group, School of Computing, University of Glamorgan, Pontypridd, UK. CF37 1DL. Phone +44 (0)1443 482731,
email pangel@glam.ac.uk
LYNNE BODDY1 & COLIN W. MORRIS2. Developing artificial neural networks for identification.
Simple, rapid identification techniques for biologists who are not taxonomic experts are becoming increasingly. Artificial neural networks (ANNs) are a tool with great promise, since they can cope with partially contradictory 'fuzzy' data. Further, while it is not trivial to select and optimize ANNs, their implementation does not need a taxonomic expert beyond the original determinations of example patterns upon which the system is to be trained. This paper describes how they work and how they can be developed for solving biological identification problems. ANNs are not rule based but learn/train from examples presented, hence it is not necessary to know the underlying distributions or relationships between parameters. Essentially, there are two types of training - supervised and unsupervised. With the former, data patterns of known identity are repeatedly presented to the ANN and, provided that the parameters measured are sufficiently discriminatory, the network will eventually be able successfully to identify taxa upon which it has been trained. With unsupervised training, on the other hand, patterns are presented to the network and it forms its own groupings of the data. The supervised approach is appropriate for making identifications, and the radial basis function (RBF) has proved most successful for biological identification. RBF ANNs train rapidly, make identifications rapidly and can reject taxa as unknown to the trained network. The unsupervised approach is useful for reducing dimensionality and revealing clusters in data sets, and can be used to aid obtaining training data from natural population samples.
1Cardiff School of Biosciences, University of Cardiff, Cardiff CF10 3TL, U.K., email BoddyL@cf.ac.uk;
2School of Computer Studies, University of Glamorgan, Pontypridd, CF37 1DL,U.K., email cwmorris@glam.ac.uk
LYNNE BODDY1, MALCOLM F. WILKINS1, COLIN W. MORRIS2. AimsNet: Software for artificial neural net analysis of analytical flow cytometry (AFC) data.
AFC rapidly generates large multivariate data sets. Analysis of these data sets is difficult using traditional multivariate statistical techniques, since data frequently do not conform to any theoretical data distribution. Artificial neural nets (ANNs) provide a non-parametric alternative. They can learn to discriminate different categories (taxa) of cells from AFC data, given a suitable training set of data for each category. Once trained, they can rapidly categorise data as belonging to one of the "known" categories, or as being "unknown". ANNs have successfully discriminated over seventy species of marine phytoplankton from their AFC signatures. AimsNet is a Windows™ application that brings the power of ANNs to the analysis of AFC data. With AimsNet, the user constructs a connected structure of simple data processing modules to describe the desired data processing steps. Data flow from one module to the next via directional connections. The data, in either ASCII text or FCS 2.0 formats, are read from file by a data source module. Processed data are written back to disc by a data sink module, in either listmode or summary form. Data processing modules can be used to perform data rescaling, removal of parameters, re-ordination method (principal components analysis, canonical variate analysis), cluster analysis, and pattern recognition via radial basis function ANNs. AimsNet can analyse the results of processing to generate a summary output file, for example a misidentification matrix. Although powerful, AimsNet is designed to be easy to use, even by those with limited knowledge of ANNs. Support: CEC grant MAS3-CT97-0080.
1Cardiff School of Biosciences, University of Cardiff, Cardiff CF10 3TL, U.K., email BoddyL@cf.ac.uk;
2School of Computer Studies, University of Glamorgan, Pontypridd, CF37 1DL,U.K.,
email
cwmorris@glam.ac.uk
LYNNE BODDY1, MALCOLM F. WILKINS1, COLIN W. MORRIS2. Identifying and quantifying phytoplankton populations using artificial neural network analysis of flow cytometry data.
Phytoplankton are key components of marine ecosystems which collectively fuel the marine food web; they have been implicated in climate control, and some groups form nuisance blooms. It is, therefore, crucial to be able to identify and quantify phytoplankton populations accurately, rapidly and preferably actually during survey cruises. In the past appropriate data have been obtained by microscopic analysis of samples rerturned to the laboratory. This is laborious and time-consuming and since it is performed a long time after sampling, interesting phenomena cannot be resampled or followed up directly. Analytical flow cytometry can provide signatures for micro-algal cells (at rates of about 103 cells sec-1) which allow species to be discriminated. Artificial neural networks (ANNs) are ideal for analysing the vast quantities of non-normal data produced, and can make identifications in near real-time. The approach will be illustrated using flow cytometry data on over 70 important marine phytoplankton taxa. Scaling up to cover the many hundreds of species found in the world's oceans is a non-trivial task; major progress towards this will be discussed, including rejection of unknown taxa, and combining ANNs trained to identify single species against a background of the rest.
1Cardiff School of Biosciences, University of Cardiff, Cardiff CF10 3TL, U.K., email BoddyL@cf.ac.uk;
2School of Computer Studies, University of Glamorgan, Pontypridd, CF37 1DL,U.K., email cwmorris@glam.ac.uk
Yu. F. IVLEV1, V. I. POPENKO2. STERM program for 3D reconstruction of biological objects for morphology and ultrastructural analysis.
STERM program consists of a set of the functions for 3D reconstruction on the basis of serial sections and allows to make morphological measurements on the reconstructed models. The program is intended for use on the basis of MathLab 5.2 software package. It consists of three main functional parts with independent interfaces: 1) "tracing" - for conversion of initial raster images to vector form; 2) "2D-montage" - for sections alignment; 3) "3D-montage" - 3D reconstruction and morphometry. Initial raster images can be processed in any graphic editor, e.g. Adobe Photoshop. Some features of STERM: supports BMP,TIFF, JPEG,PCX formats for initial raster images; enables 2 different ways of manually pre-aligned sections correction: (i) by calculation of main component of nodes coordinates in each section and (ii) by minimization of the sum of distances between contour centers in different sections; various ways of 3D reconstruction can be used; animated images can be created and viewed etc. Most of the functions of the program can be used both in automatic and manual modes. The program was tested in the Chair of zoology of vertebtata, Biological Department of Moscow State University. Initially designed for reconstruction of derivates of vertebrata epidermis, the program was also used for 3D reconstruction of mammalian organs using optical sections. At present, after improvement of tracing and alignment procedures the program can be successfully used for reconstruction of organelles and compartments of the cell on the basis of ultrathin sections containing many objects to be reconstructed. Grants: RFFI 00-15-97761, 99-04-48792, FRP "Integration" А0084.
1Severtsov Institute of ecology and evolution problems RAS, Leninsky av. 33, Moscow, 117071, Russia; Fax: (095)9545534; e-mail: ivlev@11.vertebra.bio.msu.ru
2Engelhardt Institute of Molecular Biology RAS, Vavilova str. 32, Moscow 119991 GSP-1, Russia
K.P.KOUTZENOGII1, N.V. VLASOVA2, L.K.TRUBINA3, AND A.P.GUK3.
Photogrammetric method used to analyze microobjects. The digital photogrammetric method is proposed for studying morphological characteristics of particles of the geometrically indeterminate form (characteristic size ranges from 5 mkm to several mm). The potentialities of the microscopic method of study of the form of such objects are rather limited whereas the photogrammetric method provides the necessary spatial characteristics from a stereopair of micropatterns. Experimental results are given for the study of the form of pollen grains of different plants and seeds (Caryophylacceae family) by their microimages.
1Institute of Chemical Kinetics and Combustion SB RAS, Novosibirsk, Russia;
2Siberian State Geodetic Academy, Novosibirsk, Russia;
3Central Siberian Botanic Garden SB RAS, Novosibirsk, Russia.
O. G. LEONOVA1, J. L. IVANOVA1, G. V. SERDYUKOV1, V. I. POPENKO1, Yu. F. IVLEV2. 3D-reconstruction of ciliate macronuclei using STERM software.
Each ciliate cell contains nuclei of two types: transcriptionally actve polyploid somatic macronuclei, ensuring cell functioning, and inert generative micronuclei. Contrary to high molecular weight DNA of higher eukaryotes, macronuclear DNA is represented by relatively short molecules (several tens to hundreds kbp), organized into ~100-200 nm chromatin bodies similar to chromomeres of higher eukaryotes. The aim of this work was to ascertain a spacial arrangement of chromatin bodies in different parts of a macronucleus and to study dinamic possibilities of macronuclear chromomeres. This problem is important both for elucidation of spatial organization of ciliate macronuclei and for understanding of general structural mechanisms enabling selective regulation of large groups of genes in the nucleus. We used serial ultrathin sections (20-30 sections, 50-80 nm thick) for 3D reconstruction of chromatin structures in the macronuclei of ciliates Bursaria truncatella and Paramecium aurelia. The section were photographed in JEM-100CX electron microscope. 3D reconstruction was performed on scanned negatives using STERM software (ivlev@11.vertebra.bio.msu.ru). The following conclusions were made: (i) chromatin bodies in the ciliate macronuclei are able to form thick fibrils similar to chromonemes in chromosomes of higher eukaryotes; (ii) during chromonema formation chromatin bodies firstly increase in size and then large chromomeres join into chromonema. The data obtained evidence for existence of specific zones in the macronucleus analogous to "chromosome territories" in higher eukaryotes nuclei. Grants: RFFI 99-04-48792, FRP "Integration" А0084.
1Engelhardt Institute of Molecular Biology RAS, Vavilova str. 32, Moscow 119991 GSP-1, Russia; Tel: (095)-1359804; Fax: (095)-1351405; e-mail: popenko@genome.eimb.relarn.ru
2Severtsov Institute of ecology and evolution problems RAS, Leninsky av. 33, Moscow, 117071, Russia
V.S. MUKHANOV1, A.M. LYAKH1, R.B. KEMP2. The Virtual Cell Project: a technology for semiautomatic phytoplankton cell reconstruction.
This paper introduces the Virtual Cell (VC) project which is designed for the needs of marine biologists and may be used in various scientific areas such as zoology, botany, etc. The VC project is based on the concept of 'virtual cell' - a parametric 3D model which imitates a shape of natural phytoplankton cell and its ornamentation. The main goals of the VC project are: (1) to develop the algorithms for semiautomatic reconstruction of microalgae using image information; (2) to design the methods for description of cell morphotypes; (3) to study intraspecific variations of cell morphology; (4) to improve the estimates of cell surface area and surface/volume ratio; (5) to create VC database. Each algal species is represented in the VC project as the base 3D model described by a set of characteristic points (CP). Simulation process is reduced to fitting the model CP pattern to the real one observed in the natural cell. Funded by INTAS grant № 99-01390.
1 Institute of Biology of the Southern Seas, National Academy of Science, Nakhimova 2, Sevastopol, 99011, Ukraine, mukhanov@ibss.iuf.net;
2 Institute of Biological Sciences, University of Wales, Aberystwyth, Wales, UK, rbk@aber.ac.uk
D. A. NIPPERESS, J. M. DANGERFIELD, C. J. ANGUS, R. A. BRAMBLE, A. J. PIK, D. L. SAUNDERS & A. J. BEATTIE. BioTrack: a system for rapid assessment and monitoring of biodiversity.
BioTrack, developed by staff at the Centre for Biodiversity & Bioresources, Macquarie University, Australia, provides rapid assessment and monitoring of biodiversity in a range of real world situations. BioTrack is a system which integrates laboratory and field protocols with 'off-the-shelf' software packages. Barcoding is used extensively in the collection and management of specimens while a specialised relational database is used to store and retrieve data. Operators using a morphospecies approach can generate species-level information by querying a database of high-quality, fully-focused images of voucher specimens. These images can also be used for remote verification by taxonomists. We demonstrate some applications of the BioTrack system in habitat assessment, regeneration and monitoring projects where we routinely process hundreds of thousands of specimens from hundreds of samples. Future prospects and applications for this system will be discussed.
Centre for Biodiversity & Bioresources, Department of Biological Sciences, Macquarie University, NSW, 2109, Australia, Phone +61 2 9850 7248, Fax +61 2 9850 9237, email dnippere@rna.bio.mq.edu.au
P.V.OZERSKI. Use of the artificial neuronal network method to identification of Bivalvia species by the form of their shells.
Bivalvian molluscs grow shell wall in such way, that the contour of the external border of its cross-sections can be described by a logarithmic spiral equation. The parameters of this equation are species-specific and do not change during the growth of the mollusc's body. Basing on this feature, a development of software package for IBM-compatible computers (Win32 platform) to identification of Bivalvian molluscs species by scanned images of cross profiles of theirs shells is started. By now a part of preliminary image processing including the selecting of shell contour and determining its longitudinal axis is complete. In the future the automatic determination of mollusc's body side, image orientation and the starting point of logarithmic spiral as well as the calculation of spiral parameters will be realized. To solve a part of thiese probems self-learning neuronal networks will be used. This project is supported by INTAS foundation (97-30950).
Sechenov Institute of Evolutionary Physiology and Biochemistry, 194223, St.Petersburg, M.Thores pr., 44.
I.Ya. PAVLINOV. Geometric morphometrics, a new analytical approach to comparison of digitized images.
The method of geometric morphometrics is designed for strictly numerical comparison of digitized images of morphological objects by their shapes, the effect of size factor being excluded. It is based on the concept of multidimensional "shape space" which properties are defined by so called "procrustes metrics". The objects are described not by linear measurements between landmarks but by Cartesian coordinates of these landmarks. Distribution of shapes in the shape space is described by "shape variables", among which so called "relative warps" (an analogy of principal components) are biologically most significant. The shapes, after their alignment, are compared by either "superposition" method or "resistant fit" method". Numerical evaluation of pairwise differences among individual shapes is provided by so called "procrustes distance" (an analogy of Euclidian distance possessing metric properties). The method of "deformational grid" (an analytical resolution for D'Arcy Thompson's idea) is used for visualizing these differences. Possibilities of the method are limited by initial assumption about structural homogeneity of the shapes and about continuity of their transformations. Besides, it lacks by now a numerical measure (an analogy of the coefficient of variation) of magnitude of overall shape variability. A set of computer programs is available to fulfill the geometric morphometric tasks. Use of the method and operations with the programs will be illustrated by analyses of skull and dentition shapes in some rodents.
Zoological Museum, Moscow M.V.Lomonosov State University, ul. B. Nikitskaya 6, 103009 Moscow, Russia, Phone.: (095) 203-2940, email: pvl@2.zoomus.bio.msu.ru
__________________________________________________________
Web-page prepared by
Zoological Institute RAS &
BioNET International
Last web-page update: 3rd April 2001