Designing a pattern recognition system on GPU for discriminating between patients with micro-ischaemic and multiple sclerosis lesions, using MRI images

  • Authors:

  • Ekaterini Solomou;Spiros Kostopoulos;Konstantinos Sidiropoulos;Emmanouil Athanasiadis;Eleftherios Lavdas;Dimitris Glotsos;George Sakellaropoulos;Petros Zampakis;John Stonham;Dionisis Cavouras

  • Affiliations:

  • Department of Radiology, University Hospital of Patras, Rio, Greece;Medical Image and Signal Processing Laboratory, Department of Medical Instruments Technology, Technological Educational Institute of Athens, Egaleo, Athens, Greece;School of Engineering and Design, Brunel University West London, Uxbridge, Middlesex, UK;Medical Image and Signal Processing Laboratory, Department of Medical Instruments Technology, Technological Educational Institute of Athens, Egaleo, Athens, Greece;Department of Radiologic Technologists, Technological Educational Institute of Athens, Egaleo, Athens, Greece;Medical Image and Signal Processing Laboratory, Department of Medical Instruments Technology, Technological Educational Institute of Athens, Egaleo, Athens, Greece;Department of Medical Physics, School of Medicine, University of Patras, Rio, Patras, Greece;Department of Radiology, University Hospital of Patras, Rio, Greece;School of Engineering and Design, Brunel University West London, Uxbridge, Middlesex, UK;Medical Image and Signal Processing Laboratory, Department of Medical Instruments Technology, Technological Educational Institute of Athens, Egaleo, Athens, Greece

  • Venue:
  • International Journal of High Performance Computing Applications
  • Year:
  • 2013

Quantified Score

Hi-index 0.00

Visualization

Abstract

The aim of this study was to employ state-of-art graphics processing unit (GPU) technology and CUDA parallel programming to design and implement a stand-alone pattern recognition (PR) system to discriminate between patients with micro-ischaemic (mIS) and multiple sclerosis (MS) lesions. The dataset comprised MRI image series of 32 patients with mIS and 19 with MS lesions. The probabilistic neural network classifier and 40 textural features, calculated from lesions in the magnetic resonance imaging (MRI) images, were used to design the PR system. The highest classification accuracy was 90.2%, employing six textural features. It took about 135 minutes to design the PR system on a desktop CPU (Intel Core 2 Quad Q9550), using sequential programming, against 250 seconds on the Nvidia 8800GT GPU card, using parallel programming. The proposed PR system may be redesigned on site, when new verified data are incorporated in its depository, and it may serve as a second opinion tool in a clinical environment.