Distributed Computation of Feature-Detectors for Medical Image Processing on GPGPU and Cell Processors

Abstract

Automated classification of medical (computed tomography) images may ultimately lead to faster and improved diagnosis, benefiting both patients and clinicians. We describe a software system, that can be trained for classification purposes in the area of medical image processing. The underlying algorithm is based on a set of perceptron-like feature detectors, which are combined to short feature vectors. Those are used to form self-organized Kohonen maps, which will be used for the classification of new image data. The exact description of the feature detectors is derived from a large set of sample images by way of an evolutionary strategy. This leads to a computationally demanding process of iterated image decomposition, Kohonen map training and quality assessment. To make our method feasible, we rely on clusters of rather cheap commodity hardware, namely general purposes graphics processing units (GPGPU ?) and the STI Cell Broadband Engine Architecture (Cell), as it comes with the PS3 gaming console.

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.