Improving Dose-Response Correlations For Locally Advanced Nsclc Patients Treated With Imrt Or Pspt

Author

Yulun HeFollow

Author ORCID Identifier

0000-0001-6258-4986

Date of Graduation

5-2023

Document Type

Dissertation (PhD)

Program Affiliation

Medical Physics

Degree Name

Doctor of Philosophy (PhD)

Advisor/Committee Chair

Kristy Brock

Committee Member

Radhe Mohan

Committee Member

Carlos Cardenas

Committee Member

Laurence Court

Committee Member

Zhongxing Liao

Committee Member

Carol Wu

Abstract

The standard of care for locally advanced non-small cell lung cancer (NSCLC) is concurrent chemo-radiotherapy. Despite recent advancements in radiation delivery methods, the median survival time of NSCLC patients remains below 28 months. Higher tumor dose has been found to increase survival but also a higher rate of radiation pneumonitis (RP) that affects breathing capability. In fear of such toxicity, less-aggressive treatment plans are often clinically preferred, leading to metastasis and recurrence. Therefore, accurate RP prediction is crucial to ensure tumor coverage to improve treatment outcome. Current models have associated RP with increased dose but with limited accuracy as they lack spatial correlation between accurate dose representation and quantitative RP representation. These models represent lung tissue damage with radiation dose distribution planned pre-treatment, which assumes a fixed patient geometry and inevitably renders imprecise dose delivery due to intra-fractional breathing motion and inter-fractional anatomy response. Additionally, current models employ whole-lung dose metrics as the contributing factor to RP as a qualitative, binary outcome but these global dose metrics discard microscopic, voxel-(3D pixel)-level information and prevent spatial correlations with quantitative RP representation.

To tackle these limitations, we developed advanced deformable image registration (DIR) techniques that registered corresponding anatomical voxels between images for tracking and accumulating dose throughout treatment. DIR also enabled voxel-level dose-response correlation when CT image density change (IDC) was used to quantify RP. We hypothesized that more accurate estimates of biologically effective dose distributions actually delivered, achieved through (a) dose accumulation using deformable registration of weekly 4DCT images acquired over the course or radiotherapy and (b) the incorporation of variable relative biological effectiveness (RBE), would lead to statistically and clinically significant improvement in the correlation of RP with biologically effective dose distributions.

Our work resulted in a robust intra-4DCT and inter-4DCT DIR workflow, with the accuracy meeting AAPM TG-132 recommendations for clinical implementation of DIR. The automated DIR workflow allowed us to develop a fully automated 4DCT-based dose accumulation pipeline in RayStation (RaySearch Laboratories, Stockholm, Sweden). With a sample of 67 IMRT patients, our results showed that the accumulated dose was statistically different than the planned dose across the entire cohort with an average MLD increase of ~1 Gy and clinically different for individual patients where 16% resulted in difference in the score of the normal tissue complication probability (NTCP) using an established, clinically used model, which could qualify the patients for treatment planning re-evaluation. Lastly, we associated dose difference with accuracy difference by establishing and comparing voxel-level dose-IDC correlations and concluded that the accumulated dose better described the localized damage, thereby a closer representation of the delivered dose. Using the same dose-response correlation strategy, we plotted the dose-IDC relationships for both photon patients (N = 51) and proton patients (N = 67), we measured the variable proton RBE values to be 3.07–1.27 from 9–52 Gy proton voxels. With the measured RBE values, we fitted an established variable proton RBE model with pseudo-R² of 0.98. Therefore, our results led to statistically and clinically significant improvement in the correlation of RP with accumulated and biologically effective dose distributions and demonstrated the potential of incorporating the effect of anatomical change and biological damage in RP prediction models.

Keywords

medical physics, radiation oncology, non small cell lung cancer, radiation therapy, proton therapy, computed tomography, normal tissue complication probability, deformable image registration, dose accumulation, relative biological effectiveness