Show simple item record

dc.contributor.authorAlavi, Azadehen_US
dc.contributor.authorCavanagh, Brentonen_US
dc.contributor.authorTuxworth, Gervaseen_US
dc.contributor.authorMeedeniya, Adrianen_US
dc.contributor.authorBlumenstein, Michaelen_US
dc.contributor.authorMackay-Sim, Alanen_US
dc.description.abstractKey Words: Neural Network, Classification, Three Dimensional Data, Central Nervous System. Background - Accurate three-dimensional images of neurons labelled using multiple markers can be readily generated. Such images, obtained using fluorescent markers specific to the cell type(s) of interest, were used in a heterogeneous population of cells for the accurate analysis of cell numbers [1]. Importantly, neural cells are highly plastic, mirroring their functional state in their morphology. Therefore, data from further classification of the three dimensional images from individual cells would enable the functional state of the cell to be established, including the morphological change associated with cellular dysfunction, degeneration and disease. The detection of morphological changes may be performed manually, but is open to bias and is time consuming. The availability of accurate three dimensional data in a digital medium has allowed the use of neural networks to rapidly process large data sets by 'learning' or adapting. In this paper we demonstrate using a neural network to classify 3 subpopulations of dopaminergic neurons based on their three dimensional morphological features. Methods - High resolution 3D image data from three populations of fluorescently labelled dopaminergic neurons were captured on a Zeiss axioimager with ApoTome and the resultant images processed using Imaris software by Bitplane. A number of features were extracted from each cell and the data was fed into a neural network which automatically classified each feature set as a specific cell type. The cell types were defined during the training process which involved exposing the neural network to features generated from cells of known phenotype as initially defined by their neurochemical code. Thus, the neural network identified each 'typical' cell type based on its morphological features. Results - On completion of training, the neural network was able to distinguish the three cell types to within 91% accuracy. It outperformed a human expert in accuracy (72%) and speed on the same set of data. Conclusion - With the ever increasing dimensionality and quantity of image data, information analysis is challenging to a human, thus automated methods of analysis are required. Neural networks are one possible solution that enable not only high throughput analysis, but also the adaptability inherent in the networks themselves. The unbiased automation of cell classification has multiple applications including disease diagnosis and biomedical research where the plasticity of neural morphology allows the morphometric analysis of cellular state. [1]Cavanagh, B. , Meedeniya, A.C.B. , Muller, D. , Blumenstein, M. , Mackay-Sim, A. (2007) Introducing "Fluorescence Neurosteriology": Novel methods for mapping the brain. Focus on Microscopy 2007.en_US
dc.publisherNo data provideden_US
dc.relation.ispartofconferencenameFocus On Microscopy 2009en_US
dc.relation.ispartofconferencetitleProceedings, Focus On Microscopy 2009en_US
dc.relation.ispartoflocationKrakow, Polanden_US
dc.subject.fieldofresearchNeurosciences not elsewhere classifieden_US
dc.titleNeural networks in the automated classification of neural cell morphology – rapid, unbiased morphometryen_US
dc.typeConference outputen_US
dc.type.descriptionE3 - Conference Publications (Extract Paper)en_US
dc.type.codeE - Conference Publicationsen_US
gro.facultyGriffith Sciences, School of Information and Communication Technologyen_US
gro.hasfulltextNo Full Text

Files in this item


There are no files associated with this item.

This item appears in the following Collection(s)

  • Conference outputs
    Contains papers delivered by Griffith authors at national and international conferences.

Show simple item record