Now showing 1 - 7 of 7
  • Publication
    Synthetic Positron Emission Tomography Using Conditional-Generative Adversarial Networks for Healthy Bone Marrow Baseline Image Generation
    A Conditional-Generative Adversarial Network has been used for a supervised image-to-image translation task which outputs a synthetic PET scan based on real patient CT data. The network is trained using only data of patients with healthy bone marrow metabolism. This allows for a patient specific synthetic healthy baseline scan to be produced. This can be used by a clinician for comparison to real PET data in the absence of a baseline scan or to aid in the diagnosis of conditions such as Multiple Myeloma which manifest as changes in bone marrow metabolism
  • Publication
    Machine Learning in Prediction of Prostate Brachytherapy Rectal Dose Classes at Day 30
    (Irish Pattern Recognition & Classification Society, 2015-08-28) ; ; ; ; ;
    A retrospective analysis of brachytherapy implant data was carried out on 351 patients that underwent permanent I125 brachytherapy for treatment of low-risk prostate cancer. For each patient, the dose received by 2cm3 of the rectum (D2cc) 30 days post implant was defined as belonging one of two classes, ”Low” and ”High” depending on whether or not it was above or below a particular dose threshold. The aim of the study was to investigate the application of a number of machine learning classification techniques to intra-operative implant dosimetry data for prediction of rectal dose classes determined 30 days post implant. Algorithm performance was assessed in terms of its true and false positive rates and Receiver Operator Curve area based on a 10-fold cross validation procedure using Weka software. This was repeated for a variety of dose class thresholds to determine the point at which the highest accuracy was achieved. The highest ROC areas were observed at a threshold of D2cc = 90 Gy, with the highest area achieved by Bayes Net (0.943). At more clinically useful thresholds of D2cc = 145 Gy, classification was less reliable, with the highest ROC area achieved by Bayes Net (0.613).
  • Publication
    Suppression of the CT Beam Hardening Streak Artifact Using Predictive Correction on Detector Data
    (Horizon Research Publishing Corporation, 2016-03) ;
    The purpose of the research was to develop an automated program incorporating a predictive artifact correction technique (PACT) to correct for the signal deviations from metal beam hardening artifacts in Computed Tomography (CT) detector raw data. Thin-slice sequential CT scans were performed on a dosimetry head phantom using a Somatom Sensation 16 scanner to establish a ground truth image. Metal pins were then affixed to either side of the phantom at the three and nine o'clock positions to cause streak artifact in detector raw data and a subsequent streak image. The program automatically detected the extent of the overlap peaks in the detector raw data causing the artifact. It profiled a correction using adjacent projections so that the peak error could be corrected rather than simply being removed or smoothed by interpolation. The PACT algorithm modified raw data was then reconstructed on a SYNGO CT reconstruction workstation. This image was then compared against ground truth and that produced by commercially available metal artifact reduction projection completion and also a research based iterative technique. Qualitative results illustrate superior suppression of streak artifact in images using PACT when compared directly to tested projection completion methods but inferior to iterative reconstruction. Recovery of voxel data underlying the streak is also demonstrated to be quantitatively superior with PACT when referenced to the original ground truth image. Limitations were however detected with the threshold technique for initial localisation of the streak sources. The work still demonstrates the feasibility of this predictive artifact correction technique in correcting beam hardening affected voxel data without recourse to expensive additional options such as iterative reconstruction or dual energy that are not so commonly available in the clinical setting.
  • Publication
    Anatomically accurate model of EMG during index finger flexion and abduction derived from diffusion tensor imaging
    This study presents a modelling framework in which information on muscle fiber direction and orientation during contraction is derived from diffusion tensor imaging (DTI) and incorporated in a computational model of the surface electromyographic (EMG) signal. The proposed model makes use of the principle of reciprocity to simultaneously calculate the electric potentials produced at the recording electrode by charges distributed along an arbitrary number of muscle fibers within the muscle, allowing for a computationally efficient evaluation of extracellular motor unit action potentials. The approach is applied to the complex architecture of the first dorsal interosseous (FDI) muscle of the hand to simulate EMG during index finger flexion and abduction. Using diffusion tensor imaging methods, the results show how muscle fiber orientation and curvature in this intrinsic hand muscle change during flexion and abduction. Incorporation of anatomically accurate muscle architecture and other hand tissue morphologies enables the model to capture variations in extracellular action potential waveform shape across the motor unit population and to predict experimentally observed differences in EMG signal features when switching from index finger abduction to flexion. The simulation results illustrate how structural and electrical properties of the tissues comprising the volume conductor, in combination with fiber direction and curvature, shape the detected action potentials. Using the model, the relative contribution of motor units of different sizes located throughout the muscle under both conditions is examined, yielding a prediction of the detection profile of the surface EMG electrode array over the muscle cross-section.
      213Scopus© Citations 15
  • Publication
    Cross-Correlation Template Matching for Liver Localisation in Computed Tomography
    Many of the current approaches to automatic organ localisation in medical imaging require a large amount of labelled patient data to train systems to accurately identify specific anatomical features. CrossCorrelation, also known as template matching, is a statistical method of assessing the similarity between a template image and a target image. This method has been modified and presented here to localize the liver in Computed Tomography volume images in the Coronal and Sagital planes to achieve a mean positioning error of approximately 11 mm and 20 mm respectively based on between 1 and 25 datasets to create the template liver.
  • Publication
    Applying DTI white matter orientations to finite element head models to examine diffuse TBI under high rotational accelerations
    The in-vivo mechanical response of neural tissue during impact loading of the head is simulated using geometrically accurate finite element (FE) head models. However, current FE models do not account for the anisotropic elastic material behaviour of brain tissue. In soft biological tissue, there is a correlation between internal microscopic structure and macroscopic mechanical properties. Therefore, constitutive equations are important for the numerical analysis of the soft biological tissues. By exploiting diffusion tensor techniques the anisotropic orientation of neural tissue is incorporated into a non-linear viscoelastic material model for brain tissue and implemented in an explicit FE analysis. The viscoelastic material parameters are derived from published data and the viscoelastic model is used to describe the mechanical response of brain tissue. The model is formulated in terms of a large strain viscoelastic framework and considers non-linear viscous deformations in combination with non-linear elastic behaviour. The constitutive model was applied in the University College Dublin brain trauma model (UCDBTM) (i.e. three-dimensional finite element head model) to predict the mechanical response of the intra-cranial contents due to rotational injury.
      593Scopus© Citations 43
  • Publication
    Deep Evolution of Feature Representations for Handwritten Digit Recognition
    A training protocol for learning deep neural networks, called greedy layer-wise training, is applied to the evolution of a hierarchical, feed-forward Genetic Programming based system for feature construction and object recognition. Results on a popular handwritten digit recognition benchmark clearly demonstrate that two layers of feature transformations improves generalisation compared to a single layer. In addition, we show that the proposed system outperforms several standard Genetic Programming systems, which are based on hand-designed features, and use different program representations and fitness functions.
      182Scopus© Citations 10