Speech and Palate Imaging

Our research programme in the development and implementation of MRI protocols for the management of cleft palate patients is funded by Barts and the London Charity and CLEFT

Adaptive Averaging

As well as developing the acquisition of images, work has been undertaken to increase the quality of images after acquisition. The following two movies demonstrate the improvement in image quality obtained when using a technique known as adaptive averaging. The technique has been implemented and modified for use in dynamic studies of the soft-palate. The subject repeats the test phrase several times and the images where the palate is in the same configuration are automatically detected and averaged to improve image quality. In contrast to gated techniques, the requirements on the similarity and timing of speech between repeats are minimal.

Image data acquired:

Image data after adaptive averaging processing:

The averaging technique is summarised in the following diagram:


(1) Data is acquired in real-time. (2) A region of interest (ROI) is defined to cover the soft-palate and its expected range of motion. (3) The ROI is copied to all other images. (4) The data in the ROI in each image is compared to the data in the ROI in every other image using a technique known as “cross-correlation”. The result is a matrix of cross correlation values, representing how similar the data in each image is. (5) In each row the n highest values are found, corresponding to the most similar images. (6) These values are averaged (or summed) and the result (7) is images with higher quality in the region around the soft-palate. The efficacy of the technique is further demonstrated in the next figure. When increasing the number of averages the image quality improves (particularly in the zoomed images) and intensity profile plots through the soft palate becomes much clearer:

For further details:

Scott AD , Boubertakh R , Birch MJ, Miquel ME (2012) Adaptive averaging applied to dynamic imaging of the soft palate Magnetic Resonance in Medicinedoi 10.1002/mrm.24503.