Author ORCID Identifier

0000-0003-3416-2924

Date of Graduation

5-2024

Document Type

Thesis (MS)

Program Affiliation

Medical Physics

Degree Name

Masters of Science (MS)

Advisor/Committee Chair

Laurence Court

Committee Member

Carlos Eduardo Cardenas

Committee Member

Julianne Pollard-Larkin

Committee Member

Steven Frank

Committee Member

David T Fuentes

Committee Member

Falk Poenisch

Committee Member

Zhiqian H Yu

Abstract

Radiation treatment planning is a crucial and time-intensive process in radiation therapy. This planning involves carefully designing a treatment regimen tailored to a patient’s specific condition, including the type, location, and size of the tumor with reference to surrounding healthy tissues. For prostate cancer, this tumor may be either local, locally advanced with extracapsular involvement, or extend into the pelvic lymph node chain. Automating essential parts of this process would allow for the rapid development of effective treatment plans and better plan optimization to enhance tumor control for better outcomes.

The first objective of this work, to automate the treatment planning process, was the automatic segmentation of critical structures. Delineation of both target and normal tissue structures was necessary to establish the foundation for identifying where radiation must be delivered and what should be spared from excess radiation.

Deep learning segmentation models were developed from retrospective CT simulation imaging data and clinical contours to delineate intact, postoperative, and nodal treatment structures for prostate cancer to accomplish this objective. Quality contours were extracted per established contouring guidelines in the literature. Model refinement on a holdout fine-tune dataset was used to verify model contours before quantitative and qualitative evaluation on the holdout test set. Predicted contours resulted in contours comparable in quantitative Dice-Similarity-Coefficient (DSC) and 95% Hausdorff Distance (HD95) to proposed models in literature and clinically usable contours with no more than minor edits upon physician review.

The second objective was the automation of Volumetric Modulated Arc Therapy (VMAT) planning for a breadth of prostate treatment scenarios. Development of VMAT plans for intact, postoperative, and nodal involvement treatment cases was necessary for the sequence in daily treatment delivery and the prospective distribution of radiation dose to target and normal tissues.

To accomplish this objective, knowledge-based planning models were separately developed to estimate patient-specific DVHs to guide plan optimization for radiation delivery. These two models were then used in this work for end-to-end testing of cases with and without lymph node involvement, including determining if the prostate target is intact or postoperative with or without treatment devices such as hydrogel spacers and rectal balloons. A sequence of iterative optimization runs was created to ensure hotspot reduction and target conformality.

The findings demonstrated that plans developed from automatically generated contours were clinically usable with minor edits for intact and postoperative treatments without lymph node involvement. For treatments with lymph node involvement, dose constraints were met for a select set of cases without excessive rectum curvature or excessive bladder descension into the postoperative treatment bed. When comparing auto-segmented to clinical contours, clinical contours experienced similar pass rates as those achieved by auto-segmented contours.

Keywords

Prostate Radiotherapy, Prostate Cancer, CT, Deep Learning, Knowledge-Based Planning, RapidPlan, Radiation Therapy, Automated Treatment Planning, Automated Contouring

Share

COinS