Multiparametric Classification of Tumor Treatment Using Ultrasound Microvascular Imaging
Abstract
Ultrasound-based microvascular imaging is a promising technique for evaluating tumor response to antiangiogenic therapy in preclinical settings. However, challenges such as tissue motion and noise can hinder the accuracy and reliability of contrast-free ultrasound imaging. Additionally, there is a lack of consensus on how to best combine different microvascular ultrasound techniques, like contrast-free and contrast-enhanced ultrasound, for detecting treatment response in cancer models. To address these challenges, this thesis proposes an optimal shrinkage singular value decomposition (SVD) based clutter filtering method. The proposed method significantly enhances visualization and microvascular quantification by increasing the signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR).
Additionally, a scalable preclinical tumor model is presented using ex-ovo chick chorioallantoic membrane (CAM) tumor model and machine learning algorithms. The model aims to classify renal cell carcinoma (RCC) tumor cell response to antiangiogenic treatment based on ultrasound microvascular and perfusion parameters. Perfusion parameters derived from optimal shrinkage SVD-based contrast-free ultrasound and statistical analysis of contrast-enhanced ultrasound, along with microvascular parameters from conventional analysis, are evaluated. Feature selection algorithm identifies the best combination of ultrasound-based perfusion parameters for classification. The study expands from using control and treatment groups of a sensitive cell line to using two different cell lines with varying sensitivity levels. The model pipeline is also tested on an independent cell line with unknown sensitivity to the machine learning model.
The results demonstrate the effectiveness of the model in studying antiangiogenic treatment response using ultrasound microvascular imaging. The newly developed analysis for contrast-free and contrast-enhanced ultrasound improves classification results, and the model performs well with a separate test set, demonstrating its generalization capabilities and robustness. Therefore, the proposed model pipeline has the potential to evaluate treatment response in other tumor cells and preclinical translation.
In summary, this thesis highlights that optimal shrinkage SVD-based clutter filtering method improves microvascular quantification and its vascular quantification parameters when used along with other contras-free and contrast-enhanced perfusion parameters improves the classification of the RCC tumor responses to antiangiogenic treatment in the proposed tumor chick CAM model. The results demonstrate the robustness of this study, and its potential for broader preclinical applications.