Publications
157 results found
Xie Z, 2016, Machine learning for efficient recognition of anatomical structures and abnormalities in biomedical images
Reed K, Gillies D, 2016, Automatic derivation of design schemata and subsequent generation of designs, AI EDAM-ARTIFICIAL INTELLIGENCE FOR ENGINEERING DESIGN ANALYSIS AND MANUFACTURING, Vol: 30, Pages: 367-378, ISSN: 0890-0604
Hira ZM, Gillies DF, 2016, Identifying Significant Features in Cancer Methylation Data Using Gene Pathway Segmentation, Cancer Informatics, Vol: 2016, Pages: 189-198, ISSN: 1176-9351
In order to provide the most effective therapy for cancer, it is important to be able to diagnose whether a patient's cancer will respond to a proposed treatment. Methylation profiling could contain information from which such predictions could be made. Currently, hypothesis testing is used to determine whether possible biomarkers for cancer progression produce statistically significant results. However, this approach requires the identification of individual genes, or sets of genes, as candidate hypotheses, and with the increasing size of modern microarrays, this task is becoming progressively harder. Exhaustive testing of small sets of genes is computationally infeasible, and so hypothesis generation depends either on the use of established biological knowledge or on heuristic methods. As an alternative machine learning, methods can be used to identify groups of genes that are acting together within sets of cancer data and associate their behaviors with cancer progression. These methods have the advantage of being multivariate and unbiased but unfortunately also rapidly become computationally infeasible as the number of gene probes and datasets increases. To address this problem, we have investigated a way of utilizing prior knowledge to segment microarray datasets in such a way that machine learning can be used to identify candidate sets of genes for hypothesis testing. A methylation dataset is divided into subsets, where each subset contains only the probes that relate to a known gene pathway. Each of these pathway subsets is used independently for classification. The classification method is AdaBoost with decision trees as weak classifiers. Since each pathway subset contains a relatively small number of gene probes, it is possible to train and test its classification accuracy quickly and determine whether it has valuable diagnostic information. Finally, genes from successful pathway subsets can be combined to create a classifier of high accuracy.
Reed K, 2016, Machine Learning Applications in Generative Design
The work in this thesis studies some of the potential applications of machine learning in the field of generative design. In particular it looks at how the design process can be automated once sufficient data about the design space has been collected and machine learning used to find the relationship between the design and its properties. The case study chosen for the work is the design of chairs.Preliminary work was done including the development of a parametric chair modelling program (ChairMaker) that can produce a wide range of chair designs and a series of simulations, including an automated ergonomic model, that were used to find fitness scores for desirablechair properties.New chair designs were then generated. Initially by using a well-established method; evolutionary design, using decision trees trained on the simulation data as the fitness function. The results were good, with many new viable chair designs produced. A new generative methodcalled the schema method was also developed. It extracts sets of constraints (called schemata) directly from the decision trees and uses these to generate new chairs. The schema method proved to be extremely efficient at finding viable chairs. Hundreds of diverse, original chairs can be produced within a few seconds. The idea of visual similarity was explored by using the schemata to measure the difference between two chairs. The results showed a remarkably high correlation between the volunteers considering the subjective nature of the task.The results demonstrate that it is possible to use simulated data and machine learning to make design decisions in generative design. We have shown this through the use of an existing algorithm and an original method. The new method is novel as it uses the learned knowledge about the design space directly to generate designs rather than using a search algorithm.
Xie Z, Gillies D, 2016, Patch forest: A hybrid framework of random forest and patch-based segmentation, Medical Imaging 2016: Image Processing, Publisher: SPIE, ISSN: 1605-7422
The development of an accurate, robust and fast segmentation algorithm has long been a research focus in medical computer vision. State-of-the-art practices often involve non-rigidly registering a target image with a set of training atlases for label propagation over the target space to perform segmentation, a.k.a. multi-atlas label propagation (MALP). In recent years, the patch-based segmentation (PBS) framework has gained wide attention due to its advantage of relaxing the strict voxel-to-voxel correspondence to a series of pair-wise patch comparisons for contextual pattern matching. Despite a high accuracy reported in many scenarios, computational efficiency has consistently been a major obstacle for both approaches. Inspired by recent work on random forest, in this paper we propose a patch forest approach, which by equipping the conventional PBS with a fast patch search engine, is able to boost segmentation speed significantly while retaining an equal level of accuracy. In addition, a fast forest training mechanism is also proposed, with the use of a dynamic grid framework to efficiently approximate data compactness computation and a 3D integral image technique for fast box feature retrieval.
Xie Z, Kitamoto A, Tamura M, et al., 2016, High-throughput mouse phenotyping using non-rigid registration and robust principal component analysis, Medical Imaging 2016: Image Processing, Publisher: SPIE, ISSN: 1605-7422
Intensive international efforts are underway towards phenotyping the mouse genome, by knocking out each of its 25,000 genes one-by-one for comparative study. With vast amounts of data to analyze, the traditional method using time-consuming histological examination is clearly impractical, leading to an overwhelming demand for some high-throughput phenotyping framework, especially with the employment of biomedical image informatics to efficiently identify phenotypes concerning morphological abnormality. Existing work has either excessively relied on volumetric analytics which is insensitive to phenotypes associated with no severe volume variations, or tailored for specific defects and thus fails to serve a general phenotyping purpose. Furthermore, the prevailing requirement of an atlas for image segmentation in contrast to its limited availability further complicates the issue in practice. In this paper we propose a high-throughput general-purpose phenotyping framework that is able to efficiently perform batch-wise anomaly detection without prior knowledge of the phenotype and the need for atlas-based segmentation. Anomaly detection is centered on the combined use of group-wise non-rigid image registration and robust principal component analysis (RPCA) for feature extraction and decomposition.
Reed K, Gillies D, 2016, ChairMaker - a parametric chair modelling program, Departmental Technical Report: 16/3, Publisher: Department of Computing, Imperial College London, 16/3
In this report we describe a new parametric modelling program that can produce3D models of chairs. The parameters used are easily understood physical traits such asback height and seat width. The nished ChairMaker program is capable of producinga vast range of semi-realistic 3D models from a set of 33 parameters. It is intendedthat these are used to explore automated chair design.
Xie Z, Kitamoto A, Tamura M, et al., 2016, NON-RIGID REGISTRATION AND ROBUST PRINCIPAL COMPONENT ANALYSIS WITH VARIATION PRIORS: A HIGH-THROUGHPUT MOUSE PHENOTYPING APPROACH, 13th IEEE International Symposium on Biomedical Imaging (ISBI), Publisher: IEEE, Pages: 1118-1122, ISSN: 1945-7928
Thomaz CE, Amaral V, Gillies DF, et al., 2016, Priori-driven dimensions of face-space: Experiments incorporating eye-tracking information, 9th Biennial ACM Symposium on Eye Tracking Research and Applications (ETRA), Publisher: ASSOC COMPUTING MACHINERY, Pages: 279-282
Gillies DF, Liu R, 2015, Overfiting in linear feature extraction for classificationof high-dimensional image data, Pattern Recognition, Vol: 53, Pages: 73-86, ISSN: 1873-5142
Overfitting has been widely studied in the context of classification and regression. In this paper, we study the overfitting in the context of dimensionality reduction. We show that the conventional wisdom of improving classification performance by maximising inter-class discrimination is not valid for high-dimensional datasets, and can lead to severe overfitting. In particular, we prove the theoretical existence of perfectly discriminative subspace projections, and show that for datasets with very high input dimensionality, inter-class discrimination should be reduced rather than maximised. This naturally leads to a simple dimensionality reduction technique, which we call Soft Discriminant Maps, which we use to show a direct relationship between the classification performance and the level of inter-class discrimination of feature extractors. Moreover, Soft Discriminant Maps consistently exhibit better classification performance than other comparable techniques.
Xie Z, Liang X, Guo L, et al., 2015, Automatic classification framework for ventricular septal defects: a pilot study on high-throughput mouse embryo cardiac phenotyping, JOURNAL OF MEDICAL IMAGING, Vol: 2, ISSN: 2329-4302
- Author Web Link
- Cite
- Citations: 6
Pui S, Minoi JL, Lim T, et al., 2015, Feature extraction and localisation using scale-invariant feature transform on 2.5D image, Pages: 179-187
The standard starting point for the extraction of information from human face image data is the detection of key anatomical landmarks, which is a vital initial stage for several applications, such as face recognition, facial analysis and synthesis. Locating facial landmarks in images is an important task in image processing and detecting it automatically still remains challenging. The appearance of facial landmarks may vary tremendously due to facial variations. Detecting and extracting landmarks from raw face data is usually done manually by trained and experienced scientists or clinicians, and the land marking is a laborious process. Hence, we aim to develop methods to automate as much as possible the process of land marking facial features. In this paper, we present and discuss our new automatic land marking method on face data using 2.5-dimensional (2.5D) range images. We applied the Scale-invariant Feature Transform (SIFT) method to extract feature vectors and the Otsu's method to obtain a general threshold value for landmark localisation. We have also developed an interactive tool to ease the visualisation of the overall land marking process. The interactive visualization tool has a function which allows users to adjust and explore the threshold values for further analysis, thus enabling one to determine the threshold values for the detection and extraction of important key points or/and regions of facial features that are suitable to be used later automatically with new datasets with the same controlled lighting and pose restrictions. We measured the accuracy of the automatic land marking versus manual land marking and found the differences to be marginal. This paper describes our own implementation of the SIFT and Otsu's algorithms, analyzes the results of the landmark detection, and highlights future work.
Xavier I, Pereira M, Giraldi G, et al., 2015, A Photo-Realistic Generator of Most Expressive and Discriminant Changes in 2D Face Images, 6th International Conference on Emerging Security Technologies (EST), Publisher: IEEE, Pages: 80-85
- Author Web Link
- Cite
- Citations: 3
Reed K, Gillies DF, 2015, Evolving Diverse Design Populations Using Fitness Sharing and Random Forest Based Fitness Approximation, 4th International Conference and 13th European Event on Evolutionary and Biologically Inspired Music, Sound, Art, and Design (EvoMUSART), Publisher: SPRINGER-VERLAG BERLIN, Pages: 187-199, ISSN: 0302-9743
- Author Web Link
- Cite
- Citations: 1
Hira ZM, Gillies DF, 2015, A Review of Feature Selection and Feature Extraction Methods Applied on Microarray Data., Adv Bioinformatics, Vol: 2015, ISSN: 1687-8027
We summarise various ways of performing dimensionality reduction on high-dimensional microarray data. Many different feature selection and feature extraction methods exist and they are being widely used. All these methods aim to remove redundant and irrelevant features so that classification of new instances will be more accurate. A popular source of data is microarrays, a biological platform for gathering gene expressions. Analysing microarrays can be difficult due to the size of the data they provide. In addition the complicated relations among the different genes make analysis more difficult and removing excess features can improve the quality of the results. We present some of the most popular methods for selecting significant features and provide a comparison between them. Their advantages and disadvantages are outlined in order to provide a clearer idea of when to use each one of them for saving computational time and resources.
Hira ZM, Trigeorgis G, Gillies DF, 2014, An Algorithm for Finding Biologically Significant Features in Microarray Data Based on <i>A Priori</i> Manifold Learning, PLOS ONE, Vol: 9, ISSN: 1932-6203
- Author Web Link
- Cite
- Citations: 6
Thomaz CE, do Amaral V, Giraldi GA, et al., 2014, Using an holistic method based on prior information to represent global and local variations on face images, Departmental Technical Report: 14/6, Publisher: Department of Computing, Imperial College London, 14/6
Faces are familiar objects that can be easily perceived and recognized by ourselves.However, the computational modeling of such apparently natural human ability remainschallenging. Recent studies in the literature have suggested that face processing is acognition task composed of configural (or global) and featural (or local) sources of information,but with controversial arguments about the combination of these two types ofinformation. In this work, we describe an holistic method that combines variance usedin Principal Component Analysis (PCA) with some prior knowledge about the underlyingvisual perception task, including systematically the global and local informationin the common multivariate representation of face images. We have showed that, withprior information, important local variations represented by principal components withsmall eigenvalues may not be discarded augmenting the classification accuracy of the firstorthogonal basis vectors. Most interestingly, PCA with prior knowledge provides a specializedfeature selection procedure, where the mapping of high-dimensional data into alower-dimensional space has been able to handle local variations and capture not only theentire facial appearance but also some sample group facial features.
Hira ZM, Gillies D, Curry E, 2014, Improving classification accuracy of response in leukaemia treatment using feature selection over pathway segmentation, Departmental Technical Report: 14/8, Publisher: Department of Computing, Imperial College London, 14/8
Motivation: Many people die every year from leukaemia. Some of them respond to treatment, and some of them not. This study investigates whether there is any relationship between response to treatment and features drawn from the measured methylation profiles of a set of patients. Such features could potentially be used to predict the outcome of a putative treatment regime.Results: Using AdaBoost with decision trees as weak classifiers, we managed to identify two pathways that affect classification of response and progression in blood cancer with 0.988 accuracy. We also identified a gene whose presence or absence from the dataset can drop classification accuracy from 0.988 to random.Conclusion: We identified one gene that with 99% accuracy can predict response to treatment. We were also able to identify a list of genes from the same dataset that can predict response with 0.94% accuracy.
Markides L, Gillies DF, 2014, IMPROVING BRAIN DECODING THROUGH CONSTRAINED AND PARAMETRIZED TEMPORAL SMOOTHING, 11th IEEE International Symposium on Biomedical Imaging (ISBI), Publisher: IEEE, Pages: 549-553, ISSN: 1945-7928
Markides L, Gillies DF, 2014, Intensity normalisation for large-scale fMRI brain decoding, 2014 INTERNATIONAL WORKSHOP ON PATTERN RECOGNITION IN NEUROIMAGING, ISSN: 2330-9989
Markides L, Gillies DF, 2014, Unsupervised metrics of brain region significance for event-related fMRI intersession experiments, 2014 INTERNATIONAL WORKSHOP ON PATTERN RECOGNITION IN NEUROIMAGING, ISSN: 2330-9989
Liu R, Gillies DF, 2013, An Estimate of Mutual Information that Permits Closed-Form Optimisation, ENTROPY, Vol: 15, Pages: 1690-1704
- Author Web Link
- Cite
- Citations: 2
Thomaz CE, Giraldi G, Costa J, et al., 2013, A Priori-Driven PCA, 11th Asian Conference on Computer Vision, Publisher: Springer verlag, Pages: 236-247
Principal Component Analysis (PCA) is a multivariate statistical dimensionality reduction method that has been applied successfully in many pattern recognition problems. In the research area of analysis of faces particularly, PCA has been used not only as a pre-processing step to produce accurate analytical model for automated face recognition systems, but also as a conceptual framework for human facecoding. Despite the well-known attractive properties of PCA, the traditional approach does not incorporate high level semantics from human reasoning which may steer its subspace computation. In this paper, we propose a method that allows PCA to incorporate such semantics explicitly. It allows an automatic selective treatment of the variables that compose the patterns of interest, performing data feature extraction anddimensionality reduction whenever some high level information in the form of labeled data are available. The method relies on spatial weights calculated, in this work, by separating hyperplanes. Several experiments using 2D frontal face images and different data sets have been carried out to illustrate the usefulness of the method for dimensionality reduction, interpretation, classification and reconstruction of face images.
Minoi JL, Thomaz CE, Gillies DF, 2012, Tensor-based multivariate statistical discriminant methods for face applications, ICSSBE 2012 - Proceedings, 2012 International Conference on Statistics in Science, Business and Engineering: "Empowering Decision Making with Statistical Sciences", Pages: 552-557
This paper describes the use of tensor-based multivariate statistical discriminant methods in three-dimensional face applications for synthesis and modelling of face shapes and for recognition. The methods could recognise faces and facial expressions, synthesize new face shapes and generate facial expressions based on the the most discriminant vectors calculated in the training sets that contain classes of face shapes and facial expressions. The strength of the introduced methods is that varying degrees of face shapes can be generated given that only a small number of 3D face shapes are available in the dataset. This framework also has the ability to characterise face variations across subjects and facial expressions. Recognition experiment was conducted using 3D face database created by the State University of New York (SUNY), Binghamton. The results have shown higher recognition rates for face and facial expression compared to the more popular eigenface techniques. The outcome of the synthesis of face shapes and facial expressions will also be presented here. © 2012 IEEE.
Markides L, Gillies DF, 2012, Towards identification and characterisation of selective fMRI feature sets using independent component analysis, Proceedings - 2012 2nd International Workshop on Pattern Recognition in NeuroImaging, PRNI 2012, Pages: 17-20
Pattern-information fMRI uses multivariate techniques for the interpretation of the various patterns that appear in the brain activity. Multi-voxel pattern analysis (MVPA) is a popular technique of pattern-information fMRI which enables the detection of sets of selective voxels that aid in the discrimination between two competing stimuli. Recently researchers have dealt with characterising the aforementioned sets of features by mapping them to primary cognitive processes instead of whole tasks. In this work, we demonstrate how Independent Component Analysis (ICA) provides a promising foundation for both the creation but also the characterisation of diverse sets of selective voxels that can be used later for the prediction of the nature of a given task. © 2012 IEEE.
Minoi J-L, Jupit AJR, Gillies DF, et al., 2012, Facial Expressions Reconstruction of 3D Faces based on Real Human Data, 2012 IEEE INTERNATIONAL CONFERENCE ON COMPUTATIONAL INTELLIGENCE AND CYBERNETICS (CYBERNETICSCOM), Pages: 185-189
- Author Web Link
- Cite
- Citations: 2
Markides L, Gillies D, 2012, On the Creation of Generic fMRI Feature Networks Using 3-D Moment Invariants, MLMI 2012
Multi-voxel pattern analysis (MVPA) is a common technique of pattern-information fMRI, which, through the process of feature selectionand subsequent classification, can aid the detection of groups of informative voxels that can be used to discriminate between competingstimuli. Networks of features have been long extracted univariately but recently researchers have turned to the development of multivariate techniques that also move from being purely mathematical, to have a more physiological meaning. In this work, we demonstrate a multivariate feature selection method that uses information encoded in the 3D spatial distribution of activated voxels at each anatomical region of the brain, in order to extract networks of informative regions that can act as generic features for running MVPA across subjects.
Liu R, Gillies DF, 2012, An eigenvalue-problem formulation to non-parametric mutual information maximation for linear dimensionality reduction, International Conference on Image Processing Computer Vision & Pattern Recognition (ICPV'12), Publisher: CSREA Press
Well-known dimensionality reduction (feature extraction) techniques, such as Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA), are formulated as eigenvalue-problems, where the required features are eigenvectors of some objective matrix. Eigenvalue-problems are theoretically elegant, and have advantages over iterative algorithms. In contrast to iterative algorithms, they can discover globally optimal features in one go, thus reducing computation times and avoiding local optima. Here we propose an eigenvalue-problem formulation for linear dimensionality reduction basedon maximising the mutual information between the class variable and the extracted features. Mutual information takes into account all moments of the input data while PCA and LDA only account for the first two moments. Our experiments show that our proposed method achieves better, more discriminative projectionsthan PCA and LDA, and gives better classification results for datasets in which each class is well-represented.
Minoi J-L, Gillies DF, Jupit AJR, 2012, Realistic facial expression synthesis of 3D human face based on real data using multivariate tensor methods, WSCG'2012, CONFERENCE PROCEEDINGS, PTS I & II, Pages: 69-77
Minoi J-L, Thomaz CE, Gillies DF, 2011, Synthesizing 3D Face Shapes Using Tensor-Based Multivariate Statistical Discriminant Methods, INFORMATICS ENGINEERING AND INFORMATION SCIENCE, PT IV, Vol: 254, Pages: 413-+, ISSN: 1865-0929
This data is extracted from the Web of Science and reproduced under a licence from Thomson Reuters. You may not copy or re-distribute this data in whole or in part without the written consent of the Science business of Thomson Reuters.