The theory of and practical consideration for ultrasound guided interventions: from phantom data to clinical studies
MetadataShow full item record
Cancer is one of the leading causes of death in Canada. Research in early detection is essential for improving survival rates. The current standards for diagnosis include physical examinations, chemical tests, and biopsy. However, these tests are inaccurate and invasive. There is a need for a more accurate and less invasive diagnostic tool. To address this, Temporal Enhanced Ultrasound (TeUS) was developed. TeUS is a method of non-invasive imaging based on the temporal response of the tissue to ultrasound irradiation. Previous studies have shown its effectiveness for in vivo and ex vivo classification of prostate cancer. Additional studies investigated the physical phenomenon of TeUS and demonstrated that the tissue response to physiological micro-vibrations recorded as a time series were the basis for tissue classification. This hypothesis was later validated through a series of simulations and tissue-mimicking phantom experiments. Despite the clinical success of TeUS, one underlying issue is having a controlled imaging environment with standard and repeatable micro-vibrations. Additionally, specialized ultrasound equipment is required for acquisition. This thesis aims at addressing these challenges. First, I introduced a new method of TeUS acquisition by incorporating changes to the imaging focal point in a time-dependent manner. I built 9 tissue-mimicking phantoms that differed in scatterer size and elasticity, collected TeUS, and used machine learning models to classify the phantoms. These results demonstrated the effectiveness of modifying the imaging focal point during acquisition for classification of phantoms. Second, I introduced two new methods of post-processing to further enhance ultrasound time series analysis. The first method is used to accommodate the changes to the imaging focal point, while the second method is a post-acquisition technique to create time series from a single ultrasound frame. These methods were evaluated using phantom experiments. Lastly, I demonstrated the feasibility of creating a time series from a single ultrasound frame using data collected from prostate cancer biopsy. A deep learning model was trained and results were compared to classification using traditional TeUS. The results obtained in this thesis may be useful for further improving the clinical translation of Temporal Enhanced Ultrasound for cancer diagnosis.