Ischemic stroke is usually triggered by the interruption of blood flow to the brain due to blockage of carotid arteries caused by atherosclerosis. Biomarkers developed from ultrasound images play important roles in measuring carotid atherosclerosis development and treatment response in longitudinal studies involving serial imaging.

A new method for assessing the effects of therapies on atherosclerosis is presented in this thesis by measuring the weighted average of carotid vessel-wall-plus-plaque thickness change (ΔVWT_Weighted). The changes of ΔVWT_Weighted and total plaque volume (TPV) were assessed in 120 patients randomized to pomegranate juice or extract versus placebo. Three-dimensional ultrasound (3DUS) images were acquired at baseline and one year after. 3D VWT maps were reconstructed from the lumen-intima and media-adventitia boundaries and then projected onto a carotid template to obtain two-dimensional VWT maps. Anatomic correspondence on the 2D VWT maps was optimized to reduce misalignment for the same subject and across subjects. A weight was computed at each point on the 2D VWT map to highlight anatomic locations likely to exhibit plaque progression or regression, resulting in ΔVWT_Weighted for each subject. TPV was measured as in previous studies from manual plaque segmentation. There was no significant difference in TPV-change between active therapy and placebo, but the weighted average of VWT-change measured from the 2D VWT maps with correspondence alignment (ΔVWT_Weighted,MDL) detected a significant difference between the pomegranate and placebo groups (P=0.008).

A novel deep neural network, called the deep discriminant network (DDN), is further proposed in this thesis to automatically extract the biomarker from VWT maps obtained from 3DUS images. The network can extract local features with better discrimination and optimize the weights assigned to each local feature. The sensitivity of the proposed biomarker was analyzed in 120 subjects involved in a placebo-controlled clinical trial of the effect of pomegranate on the development of carotid atherosclerosis. The results show that this new biomarker detected a significant difference between the pomegranate and placebo groups (P=9.34×10^-4) and the sample size required to establish significance was 57% smaller than the biomarker ΔVWT_Weighted directly calculated from VWT-change maps at the same setting.

Automated segmentation of carotid lumen-intima boundary (LIB) and media-adventitia boundary (MAB) by deep convolutional neural networks (CNN) from 3DUS images has made assessment and monitoring of carotid atherosclerosis more efficient than manual segmentation. However, CNN training still requires manual segmentation of LIB and MAB. Therefore, there is a need to improve the efficiency of manual segmentation and develop strategies to improve segmentation accuracy by the CNN for serial monitoring of carotid atherosclerosis. One strategy to reduce segmentation time is to increase the interslice distance (ISD) between segmented axial slices of a 3DUS image, while maintaining the segmentation reliability. The effect of ISD on the reproducibility of MAB and LIB segmentation was investigated in this thesis. The intra-observer reproducibility of LIB and MAB segmentation at ISDs of 1 mm and 2 mm was not statistically significantly different, whereas the reproducibility at ISD = 3 mm was statistically lower. Therefore, it is concluded that segmentation with an ISD of 2 mm provides sufficient reliability for CNN training. Furthermore, an alternative data partitioning strategy for serial monitoring studies is proposed that involves training the CNN by the baseline images of the entire cohort of patients for automatic segmentation of the follow-up images acquired for the same cohort (time-based partition). It was validated that segmentation with this time-based partitioning approach is more accurate than that produced by patient-based partitioning, especially at the carotid bifurcation. These findings form the basis for an efficient, reproducible, and accurate 3DUS workflow for serial monitoring of carotid atherosclerosis.

The biomarkers proposed in this thesis can improve the sensitivity of detecting treatment effects, and the proposed segmentation strategies allow for more efficient and accurate access to algorithm segmentation results for rapid computation of these biomarkers. This technology can be readily integrated into clinical workflows to improve the cost-effectiveness of proof-of-concept studies involving new therapies for atherosclerosis.