Segmentation of MRI Prostate Images using Gaussian Mixture Models (GMMs)

Image processing is advancing rapidly as data acquisition through digital medium is
becoming more common. The advancement of data acquisition directly affects the
medical field, specifically in terms of processing the images produced by Computed
Tomography (CT) scan, Magnetic Resonance Imaging (MRI), ultrasound, and XRay.
These wide range of imaging modalities would need a faster, more reliable
algorithm for image segmentation to cater for the different characteristics and needs
of image processing. Automated segmentation approach is extensively studied as the
conventional approach, which is analysing the images manually, has been proven to
be very time consuming besides being susceptible to human errors. In this paper, the
performance of Gaussian Mixture Model method for medical image segmentation is
assessed with the help of Expectation-Maximization (EM) method for data training
and Bayesian Information Criterion (BIC) function to prevent data overfitting. This
method is evaluated against MRI prostate images in particular. This method is also
compared to K-Means algorithm to evaluate its performance with synthetic image
and also clinical MRI image.