SU-E-T-271: Quantitative Analysis of Correlations Between Patient Geometry and Dosimetric Variation for Prostate Cancer Cases Using PCA and Step-Wise Regression

Authors


Abstract

Purpose:

To study the potential correlations between patient geometry information and corresponding plan dosimetric variation in terms of planning target volumes (PTVs) coverage and organs-at-risk (OARs) dose sparing for prostate cancer cases.

Methods:

We extracted dosimetric and geometrical data from twenty-three clinically acceptable prostate treatment plans. Dosimetric data extracted from each patient was PTV dose-volume historgram (DVH), and dose-wall-volume histograms (DWHs) for the bladder and rectum. Geometrical data extracted was the PTV volume, OAR volumes, and COP (closeness of OAR to PTV). COP is the weighted sum of the overlapping wall histogram (OWH). OWH is proposed to describe the complex spatial relationship between PTV and a hollow organ. Principal component analysis (PCA) was performed on the (inverted and downsampled) DVH and DWH data from all the patients. We then fit the patient geometry features to first 3 PCA scores using a step-wise regression analysis. The fitting results in a mapping from plan geometry information to dosimetric features.

Results:

First 3 PCA modes were found corresponded to clinically relevant features, indicating a tradeoff between homogenous coverage of the PTV and higher OAR dose, higher average dose to the OARs and PTV, and bladder and rectum dose tradeoffs along the first, second and third principal component respectively. Step-wise regression results show baldder surface volume and OAR COP more relevant to the 1st PCA modes score rather than other features with p-value of 0.58 for bladder volume, 0.26 for bladder COP, 0.98 for rectum COP, and less than 0.025 for other features after t-test.

Conclusion:

Quantitative correlations were established between patient geometry and corresponding plan dosimetric variation in terms of PTV coverage, trade-offs between PTV coverage and OAR dose sparing and trade-offs between different OAR dose sparing. Thus a new patient dosimetric variation tendency can be predicted and act as guidance for setting up planning goals.

Ancillary