Quantitative evaluation method of image segmentation techniques for Magnetic Resonance guided High Intensity Focused Ultrasound therapy

This paper describes a quantitative evaluation method for the accuracy of two different segmentation techniques for the treatment planning of Magnetic Resonance guided High Intensity Focused Ultrasound (MRgHIFU). The first technique is a combination of image segmentation methods consisting of Otsu's method, global, edge detection with Laplacian of Gaussian method, region growing algorithm and variable thresholding method. The second technique is a combination of image segmentation methods consisting of Otsu's method and the selection of regions using variable thresholding. These methods were used to classify the pixels of real Magnetic Resonance (MR) images obtained for the study of the distribution of heat in abscess treatment in a murine model with High-Intensity Focused Ultrasound (HIFU). In the evaluation, a total of nine surveys of 48 images each were used, and a methodology including three main steps was followed: establishment of ground truth images and calculation of areas from the segmented images, discrepancy measure calculation, and data normalization. For the evaluation an area-based metric was used and it was based on a discrepancy measure proposed for two regions and on the generalized version for c regions. After the evaluation of both segmentation techniques it was found that they presented a better performance in axial MR images than in sagittal MR images. In sagittal MR images, the average error and standard deviation error measures indicated a high variability in the segmentation for both techniques. Due to the performance of the segmentation for sagittal images, improvements will be implemented taking into account the combination of the evaluated methods in order to exploit the benefits of each one.