Doctor of Philosophy
Image-guided medical interventions are diagnostic and therapeutic procedures that focus on minimizing surgical incisions for improving disease management and reducing patient burden relative to conventional techniques. Interventional approaches, such as biopsy, brachytherapy, and ablation procedures, have been used in the management of cancer for many anatomical regions, including the prostate and liver. Needles and needle-like tools are often used for achieving planned clinical outcomes, but the increased dependency on accurate targeting, guidance, and verification can limit the widespread adoption and clinical scope of these procedures. Image-guided interventions that incorporate 3D information intraoperatively have been shown to improve the accuracy and feasibility of these procedures, but clinical needs still exist for improving workflow and reducing physician variability with widely applicable cost-conscience approaches. The objective of this thesis was to incorporate 3D ultrasound (US) imaging and image processing methods during image-guided cancer interventions in the prostate and liver to provide accessible, fast, and accurate approaches for clinical improvements.
An automatic 2D-3D transrectal ultrasound (TRUS) registration algorithm was optimized and implemented in a 3D TRUS-guided system to provide continuous prostate motion corrections with sub-millimeter and sub-degree error in 36 ± 4 ms. An automatic and generalizable 3D TRUS prostate segmentation method was developed on a diverse clinical dataset of patient images from biopsy and brachytherapy procedures, resulting in errors at gold standard accuracy with a computation time of 0.62 s. After validation of mechanical and image reconstruction accuracy, a novel 3D US system for focal liver tumor therapy was developed to guide therapy applicators with 4.27 ± 2.47 mm error. The verification of applicators post-insertion motivated the development of a 3D US applicator segmentation approach, which was demonstrated to provide clinically feasible assessments in 0.246 ± 0.007 s. Lastly, a general needle and applicator tool segmentation algorithm was developed to provide accurate intraoperative and real-time insertion feedback for multiple anatomical locations during a variety of clinical interventional procedures. Clinical translation of these developed approaches has the potential to extend the overall patient quality of life and outcomes by improving detection rates and reducing local cancer recurrence in patients with prostate and liver cancer.
Summary for Lay Audience
Medical procedures that use imaging are useful for diagnosis or treatment of patients as open surgery is often avoided, reducing the side-effects and time needed for healing. Needle-like tools are often used for managing diseases, like prostate and liver cancer, by taking samples for testing, bringing radiation directly into a tumor for treatment or heating small regions in the body to kill the cancer cells. Although the small tool sizes are helpful, higher physician skills are needed to read 2D images for understanding the 3D body while guiding these tools, which can lead to missed cancer diagnoses and treatments that have cancer recurrence. 3D information has been shown to reduce the occurrence of these poor procedure outcomes, but systems that generate 3D information are often expensive and make procedures longer. The goal of this work was to use 3D ultrasound (US) imaging with the software during image-guided prostate and liver cancer procedures.
One software approach was created to correct for prostate motion and performed with small errors at more than 15 times per second when installed in our 3D US system. Another software approach was made to automatically recognize the prostate in 3D US images. This method performed in less than one second and had the same error as humans when tested on images from different procedures, demonstrating the multi-purpose potential of the software. A new 3D US system was made for guiding liver cancer therapy by controlling three motors to create new types of images and a clinical navigation procedure that guided therapy needles to targets identified in the images. The software was made for this system to recognize needles in 3D US images in less than one second, improving the speed that needle placements could be checked. Lastly, the software was made to automatically recognize needles in 2D US images from a large range of clinical procedures and areas in the body. We believe that this research will increase the use of image-guided needle procedures to help patients with cancer while taking advantage of the reduced side-effects and time for healing.
Gillies, Derek J., "A Novel System and Image Processing for Improving 3D Ultrasound-guided Interventional Cancer Procedures" (2020). Electronic Thesis and Dissertation Repository. 7066.