Research Overview

Statistics and inference for medicine

The main aim in a clinical trial is is to examine and quantify the effectiveness of a treatment of interest. Effctiveness is evaluated relative to the effctiveness of a particular reference, the so-called control intervention. The form that the control intervention takes is dependent on the nature of the treatment which is studied. To ensure that the aforementioned comparison is meaningful, it is of essential importance to ensure that any factors not inherently associated with the two interventions (treatment and control) are normalized (controlled) between the two groups. This ensures that the observed differential outcome truly is the effect of differing interventions rather than any orthogonal, confounding variables.

A related challenge is that of blinding (or masking). Blinding refers to the concealment of the type of administered intervention from the individuals/patients participating in a trial and its aim is to eliminate differential placebo effect between groups. Although conceptually simple, the problem of blinding poses difficult challenges in a practical clinical setup. Consider two specific challenges which most strongly motivate my work. The first of these stems from the difficulty of ensuring that absolute blinding with respect to a particular trial variable is achieved. The second challenge arises as a consequence of the fact that blinding can only be attempted with respect to those variables of the trial which have been identified as revealing of the treatment administered. Put differently, it is always possible that a particular variable which can reveal the nature of the treatment to a trial participant is not identified by the trial designers and thus that no blinding with respect to it is attempted or achieved. This is a ubiquitous problem, present in every controlled trial, and one which can severely affect the trial's outcome.

Given that it is both practically and in principle impossible to ensure perfect blinding, the practice of assessing the level of blinding after the commencement of a trial has been gaining popularity and general acceptance by the clinical community. The key idea is to use a statistical model and the participants' responses to a generic questionnaire to quantify the participants' knowledge about the administered intervention. I argue that the currently employed approaches suffer from several important limitations:

(i) they necessitate the inclusion of ad hoc free parameters in the underlying statistical model,

(ii) the assumptions underlying the interpretation of the auxiliary questionnaire responses are not universally upheld,

(iii) the blinding assessment can be highly sensitive to small changes in the participants' questionnaire responses, and

(iv) they lead to a separation of assessment concerning the extent of trial blindness and of differential effectiveness of the treatment.

Motivated by these limitations of previous work, I proposed a novel statistical for integrated trial assessment which is experimentally shown to produce more meaningful and more clearly interpretable data. One of the key ideas is that it is meaningful to compare only the corresponding treatment and control participant sub-groups, that is, sub-groups matched by their auxiliary responses. The inference of the differential effect of treatment is then achieved through Bayesian analysis.

Selected publications:

  • O. Arandjelović. A New Framework for Interpreting the Outcomes of Imperfectly Blinded Controlled Clinical Trials. PLoS ONE, 2012. [LINK]

  • O. Arandjelović. Assessing Blinding in Clinical Trials. Neural Information Processing Systems, 2012. [LINK]

  • Modelling in physiology

    Weight or resistance training is an effective and popular method for increasing muscular strength. It has been successfully used in a wide range of populations, ranging from rehabilitation patients, adolescents, and the elderly, to endurance and strength athletes. Therefore, it is not surprising that much research effort has been devoted to studying the mechanisms of underlying adaptation.

    However, not all scientific advancements in understanding the effects of weight training are readily incorporated into widely employed training practices. While there can be little doubt that there are manifold reasons why this is the case, including limited communication between the two communities, it is likely that one of the key factors lies in the limited direct utility of many empirical findings. The number of training variables (such as the training frequency, intensity, exercise selection etc.), all of which mutually interact, is large, forcing empirical studies to confine their focus to a small subset of possible combinations which may not immediately be applicable in real training conditions.

    In a series of papers, I introduced and applied a mathematical model of adaptation to weight training which attempts to unify established physiological principles and data from empirical studies. Focusing the attention onto already skilled (trained) athletes, my aim was to capture and compactly describe those aspects of performance which are trainable, and to propose a method of estimating their response to a particular resistance training modality.

    Selected publications:

  • O. Arandjelović. A Mathematical Model of Neuromuscular Adaptation to Resistance Training and Its Application in a Computer Simulation of Accommodating Loads. European Journal of Applied Physiology, 110(3):523-538, 2010. [LINK]

  • O. Arandjelović. Optimal Effort Investment for Overcoming the Weakest Point New Insights from a Computational Model of Neuromuscular Adaptation. European Journal of Applied Physiology, 111(8):1715-1723, 2011. [LINK]

  • O. Arandjelović. Does Cheating Pay: The Role of Externally Supplied Momentum on Muscular Force in Resistance Exercise. European Journal of Applied Physiology, 2012. [LINK]

  • Face recognition using appearance manifolds

    Automatic face recognition has long been established as one of the most active research areas in computer vision. In spite of the large number of developed algorithms, its real-world performance has been, to say the least, disappointing. Even in very controlled imaging conditions, such as those used for passport photographs, the error rate is as high as 10%, while in less controlled environments the performance degrades even further.

    The goal of this work is to use video to achieve greater robustness of face recognition by resolving some of the inherent ambiguities (shape, texture, illumination etc.) of single-shot recognition.

    Indeed, the nature of many practical applications is such that more than a single image of a face is available. In surveillance, for example, the face can be tracked to provide a temporal sequence of a moving face. For access-control use of face recognition the user may be assumed to be cooperative and hence be instructed to move in front of a fixed camera. This is important as a number of technical advantages of using video exist: person-specific dynamics can be learnt, or more effective face representations be obtained (e.g. super-resolution images or a 3D face model) than in the single-shot recognition setup.

    Selected publications:

  • O. Arandjelović and R. Cipolla. Achieving Robust Face Recognition from Video by Combining a Weak Photometric Model and a Learnt Generic Face Invariant. Pattern Recognition, 46(1):9-23, 2013. [LINK]

  • O. Arandjelović and R. Cipolla. Computationally Efficient Application of the Generic Shape-Illumination Invariant to Face Recognition from Video. Pattern Recognition, 45(1):92-103, 2012. [LINK]

  • Face recognition-based video retrieval and organization

    In most cases humans are at the centre of interest in video. Our aim of this research is to retrieve, and rank by confidence, shots based on the presence of specific persons. Possible applications include:

    DVD browsing: Current DVD technology allows users to quickly jump to the chosen part of a film using an on-screen index. However, the available locations are predefined. Our technology could allow the user to rapidly browse scenes by formulating contextual queries.

    Content-based web search: Many web search engines have very popular image search features (e.g. Currently, search is performed based on the keywords that appear in picture filenames or in the surrounding web page content. By focusing on the content of images, the retrieval can be made much more accurate.

    Our approach consists of computing a numerical value, a distance, expressing the degree of belief that two face images belong to the same person. Low distance, ideally zero, signifies that images are of the same person, whilst a large one signifies that they are of different people.

    The method involves computing a series of transformations of the original image, each aimed at removing the effects of a particular extrinsic imaging factor. The end result is a signature image of a person, which depends mainly on the person's identity (and expression) and can be readily classified.

    Selected relevant publications:

  • O. Arandjelović and A. Zisserman. Automatic Face Recognition for Film Character Retrieval in Feature-Length Films. In Proc. IEEE Conference on Computer Vision and Pattern Recognition, Vol. 1, pages 860-867, 2005. [LINK]

  • O. Arandjelović and R. Cipolla. Automatic Cast Listing in Feature-Length Films with Anisotropic Manifold Space. In Proc. IEEE Conference on Computer Vision and Pattern Recognition, Vol. 2, pages 1513-1520, 2006. [LINK]

  • Miscellaneous problem in face recognition (Representation, Matching...)

    Selected publications:

  • O. Arandjelović. Colour Invariants Under a Non-Linear Photometric Camera Model and Their Application to Face Recognition from Video. Pattern Recognition, 45(7):2499-2509, 2012. [LINK]

  • O. Arandjelović. Gradient Edge Map Features for Frontal Face Recognition Under Extreme Illumination Changes. British Machine Vision Conference, 2012. [LINK]

  • O. Arandjelović. Recognition from Appearance Subspaces Across Image Sets of Variable Scale. British Machine Vision Conference, 2010. [LINK]

  • Multi-modal face biometrics

    Variations in head pose and illumination are some of the most practically challenging aspects of face recognition. The effects of changing pose are usually less problematic and can oftentimes be overcome by acquiring data over a time period e.g. by tracking a face. In contrast to pose, illumination changes are much more difficult to deal with: the illumination setup in which a face is imaged is in most cases not possible to control, its physics difficult to accurately model and training data containing typical appearance variability is usually not available.

    Thermal spectrum imagery is useful in this regard as it is virtually insensitive to illumination changes. On the other hand, it lacks much of the individual, discriminating facial detail contained in visual images. In this sense, the two modalities can be seen as complementing each other. The key idea is that robustness to extreme illumination changes can be achieved by fusing the two modalities.

    Selected relevant publications:

  • O. Arandjelović, R. I. Hammoud and R. Cipolla. Thermal and Reflectance Based Personal Identification Methodology in Challenging Variable Illuminations. Pattern Recognition, 43(5):1801-1813, 2010. [LINK]

  • Statistics and machine learning

    Incremental learning of GMMs is a surprisingly difficult task. One of the main challenges of this problem is the model complexity selection which is required to be dynamic by the very nature of the incremental learning framework. Intuitively, if all information that is available at any time is the current GMM estimate, a single novel point never carries enough information to cause an increase in the number of components.

    We define and consider a special, but particularly common and useful class of Gaussian mixtures - temporally coherent GMMs. Unlike previous approaches which universally assume that new data comes in blocks, each representable by a GMM, this allows our method to perform well also in the important case when novel data points arrive one-by-one, while requiring little additional memory. The key concept is that of "Historical GMM", which is the oldest GMM fit of the same complexity as the current one.

    Selected relevant publications:

  • O. Arandjelović and R. Cipolla. Incremental Learning of Temporally-Coherent Gaussian Mixture Models. In Proc. IAPR British Machine Vision Conference, Vol. 2, pages 759-768, 2005. [LINK]

  • Semi-supervised 3D segmentation of MRI images

    Segmentation of medical images is very common and useful procedure. It can be directly used for volumetric estimates, or as a preprocessing step before further analysis of data, such as morphological abnormality recognition. Hence, there is a great demand for fast and accurate segmentation.

    However, inherent underlying problems contained in this task, despite continuous efforts put into solving it, still mean that fully automatic segmentation has not been achieved. Imaging methods such magnetic resonance imaging (MRI) or ultrasound scans typically produce low contrast or noisy images, with sometimes anisotropic distortions, while scanned tissues exhibit extraordinary variability in density and shape. Our work employs statistical and inference techniques to learn on the fly from minimal user feedback, rapidly producing robust 3D segmentation results.

    Selected relevant publications:

  • O. Arandjelović. Live-Wire 3D Medical Images Segmentation. BA Dissertation, Department of Engineering, University of Oxford, 2002. [LINK]