Our current efforts are directed in three key areas: developing alternative training strategies and generalising the class of motions that the tracking algorithm can handle; improving the detection of image features; and developing further insight into the clinical interpretation of the deformation parameters.
In Section
2
we outlined how the tracking algorithm is trained using a single-step
estimation of the system matrices
and
. We are currently investigating how we could use a related idea to
build a generic model of heart motion from training data based on an
average model. We plan to investigate how well this type of model can
represent the dynamics of a normal heart and different heart conditions.
Clearly the general model is not going to track a specific heart as well
as a tracking algorithm tuned specifically to an individual heart.
However, the goal here is to provide a general enough description of
heart dynamics that can be used in conjunction with a robust feature
detector to provide robust tracking results.
It is clear that the main way in which tracking performance can be further improved is through the development of new methodology for robust acoustic boundary feature measurement. In Section 4 we found that spatio-temporal noise reduction improved image feature detection. We plan to investigate methods, possibly based on energy filters [ 9 ] (wavelets) and anisotropic diffusion [ 15 , 17 ], to extend the idea of spatio-temporal acoustic boundary detection to a truly 3D (2D+T) filtering process.
Finally, the ultimate measure of the success of this work will be to demonstrate that it is possible to relate the tracking parameters to clinical meaningful descriptors of the cardiac performance. We plan to evaluate the clinical potential of our algorithms using the objective quantification of ischemic heart disease and stress testing as example cardiac application domains.
Gary Jacob