This project is developing image synthesis techniques that recover images with both the desired tissue contrast and a normalized intensity.
The performance of image analysis algorithms applied to magnetic resonance images is strongly influenced by the pulse sequences used to acquire the images. Algorithms are typically optimized for a targeted tissue contrast obtained from a particular implementation of a pulse sequence on a specific scanner. There are many practical situations, including multi-institution trials, rapid emergency scans, and scientific use of historical data, where the images are not acquired according to an optimal protocol or the image with the desired tissue contrast is entirely missing. This project is developing image synthesis techniques that recover images with both the desired tissue contrast and a normalized intensity profile. To synthesize images, we are using image patches as features with machine learning techniques applied to training examples. Developed methods include sparse reconstruction, random forest regression, and maximum a posteriori estimation assuming a Gaussian mixture model. The methods have been shown to be useful in longitudinal analysis of brain changes in aging subjects, construction of average atlases of brain images in populations, attenuation correction in positron emission tomography image reconstruction, correcting bias in brain image segmentation, super-resolution, and distortion correction in diffusion magnetic resonance imaging. Current work focuses on image synthesis for improvement of multi-modal image registration.
|Figure 1. (a) Original MR coronal T1-weighted image cross section. (b) Original MR coronal T2-weighted image cross section. (c) Synthetic MR coronal T2-weighted images cross section showing higher resolution than the original acquired image.|
|Figure 2. (a) Original MR cross section of a human brain. (b) Synthetic computed tomography image of the same cross section.|