Gradient cumulative filtering to detect MRI thermometry artifacts

Magnetic resonance imaging (MRI) thermometry is applied for guiding thermal therapy of tumorous tissues using high-intensity focused ultrasound (HIFU). Fast imaging sequences are needed for the on-line MRI operation, and strong artifacts on the temperature map can be found especially for image regions having poor signal-to-noise (S/N) ratio. Detection and masking of the regions with unreliable temperature measurement would be important to avoid unnecessary notification of the HIFU operator or even cancelation of the sonication procedure. This paper presents a specific image processing procedure to improve detection of typical MRI thermometry artifacts. First a region for well behaving temperature map is selected. Then, the remaining region of distorted temperature surface is processed with a novel method using oriented filter on the surface gradient. The result gives a more stable surface over the artifact regions, which can be used further to improve consistency in detection and masking of the temperature errors. Evaluation data is from N=7 patients having thermal therapy for uterine fibroid, including about 300 different sonications and 30,000 MRI thermometry images. The results show that the suggested method decreases variance of the temperature difference between consequent image frames in comparison with a conventional method using 2D phase unwrapping.