GENETIC ALGORITHMS FOR MULTI-THRESHOLDING OPTIMIZATION FOR MRI SEGMENTATION

Resumo

Image segmentation is an essential task in image processing that aims to simplify and/or
change its representation for easier analysis. Medical images of the brain are extremely important in
the detection and diagnosis of diseases, as well as relevance in research in the biological area. In this
article, an approach for the segmentation of brain images is proposed, using the method based on
genetic algorithms with multiple thresholds, in order to find thresholds that optimally separate gray
matter, white matter and cerebrospinal fluid in cranial images. The most appropriate threshold
parameters were identified in order to obtain the best possible solution for image segmentation.
Magnetic resonance imaging of 10 patients was obtained from the National Center for Image Guided
Therapy database. Next, a preprocessing was established to improve the input images in the
segmentation algorithm. It was based on the application of multi-threshold image segmentation, which
aims to maximize intra-class variation between object(s) and background, as well as the minimization
of inter-class variation, i.e. between background pixels among pixels of objects, then obtaining the
tissues of interest. The genetic algorithm proved to be an efficient method to obtain optimal values for
the thresholds, so it is possible to perform segmentation of the image showing the brain tissues of
interest.