|M.Sc Student||Dan Yokhai|
|Subject||Method for Enhancing Tumor Vasculature Visualization in|
Contrast-Enhanced Ultrasound (CEUS) Clips
|Department||Department of Biomedical Engineering||Supervisor||PROFESSOR EMERITUS Dan Adam|
|Full Thesis text|
Monitoring the progress of vasculature angiogenesis of pancreatic tumors may lead to early diagnosis, more effective treatment and improvement of survival rates. While ultrasound fundamental harmonic imaging (FHI) (i.e. B-mode) is used to visualize a variety of tissues in the human body, angiogenesis is poorly visualized due to lack of echoes from blood and the noisy nature of ultrasound images of organs in great depth. Contrast-enhanced ultrasound (CEUS) is a modality that provides image enhancement of the vasculature, improving the diagnostic accuracy of angiogenesis. CEUS is less effective when continuous contrast agent (CA) injection is used (as in the data processed here). CEUS utilizes microbubbles that resonate when hit by ultrasound waves, specifically at the half harmonic of the fundamental transmitted frequency (sub-harmonic imaging (SHI)). In CEUS clips, acquired while continuous injection of CA is performed, often not only the vasculature is perfused, but also all the capillaries in the tissue surrounding it, which degrades the signal to noise ratio and makes it more challenging to provide a good vasculature visualization. Moreover, the poor spatial-temporal resolution of the clips furthermore worsens the ability to distinguish between different types of tissues. Improving vascular visualization by separating the vasculature from non-vascularized tissue may better allow monitoring pancreatic cancer angiogenesis monitoring due to anti-angiogenetic therapy.
In this study, CEUS spatial-temporal features were extracted from SHI clips. A two-step segmentation algorithm is proposed, analysing spatial-temporal dynamic properties of the CA movements in the CEUS clip. Four types of areas were segmented. In the first step, a coarse segmentation was performed: 1) tissue with small blood vessels - characterized by noisy areas where the CA was flowing for a long time after injection, 2) large blood vessels - characterized by local changing intensity dynamics (due to blood pulsation), 3) high concentrations of CA, resulting in a highly saturated environment. In the second step, fine segmentation was performed to segment CA groups within the tissue, which are not saturated but contain passing groups of CA, probably flowing through finer blood vessels. This segmentation process allows to estimate a segmentation map of coarse and a fine vasculature.
The method was validated on a phantom data, produced under a controlled environment. In vivo data of mice leg tumor was processed to validate the method in better conditions than the human pancreas. Mice leg suffered less from the noisy environment and it was hardly moving during the recording session.
The human pancreas data suffered from severe low SNR (the organ located deeply, and the CA was continuously injected), low spatial-temporal resolution (due to the data acquisition method) and movements (caused by patient breathing). To partially handle this issue, non-rigid registration was performed on the human data prior to the segmentation.
The results of the segmentation process applied to the phantom, mice tumor and human pancreas show good segmentation results, and provide better visualization of different vascularization regions. Monitoring over time the expansion or shrinkage of those regions may allow monitoring tumor progression.