Abstract:
Strain imaging, also known as elastography, is an emerging medical imaging modal-
ity for detection and diagnosis of pathologic soft tissue via its sti®ness measurement.
Assuming the tissue sti®ness as a continuous function in a physical proximity, a cost
function maximization based approach is adopted in this thesis to develop a gradient-
based (GBASE) and direct average strain estimation (DASE) techniques for high qual-
ity average strain imaging. The cost function is de¯ned from exponentially weighted
neighboring window pre- and post-compression radio-frequency (RF) normalized cross-
correlation (NCC) peaks in the lateral (for GBASE) or in both the axial and the lateral
(for DASE) directions to estimate an average strain. The proposed techniques are ro-
bust to decorrelation noise and have a built-in smoothing feature to ensure controlled
continuity in displacement/strain in neighborhood tissues.
It is also seen that with increased stress, the non-axial motions of tissue elements
increase that result in noisier strain images. At high strain, envelope of the RF signal
exhibits robustness to this signal decorrelation though the precision of the estimated
strain is much worse compared to that using the RF signals. In this thesis, we also
propose a novel approach for robust strain estimation by combining weighted RF NCC
and envelope NCC functions. An applied strain dependent piecewise-linear-weight is
used to combine them. In addition, we introduce non-linear di®usion ¯ltering to further
enhance the resulting strain image.
While strain continuity can be ensured using the proposed weighted nearest neigh-
bor approach or by using other regularization/smoothing based techniques, the lesion
edge blurring cannot be prevented as they are not intelligent enough to detect and
preserve edges while smoothing. In this thesis, a novel approach with built-in lesion
edge preservation technique is proposed for direct average strain imaging. An edge de-
tection scheme, conventionally used in di®usion ¯ltering is modi¯ed here for extracting
edge information while ensuring strain continuity. Based on the extracted edge infor-
mation, lesion edges are preserved by modifying the strain determining cost function
in the proposed DASE method in such a way that only the NCC peaks of similar sti®er
region to that of the interrogative tissue point are incorporated into it.
The proposed algorithms demonstrate signi¯cantly better performance, in terms
of both quantitative and qualitative indices, than the other reported strain estima-
tion techniques for a wide range of applied strain in ¯nite element modeling (FEM)
simulation, phantom experiment, and also with in vivo breast data.
