Proton radiography as a quality control tool for 3D and 4D thorax synthetic CTs

Speaker: Carmen Seller Oria

Abstract

Purpose: The suitability of synthetic CTs (sCT) for proton dose calculations depends, among other factors, on the Hounsfield Unit (HU) accuracy. The purpose of this study was to evaluate 3D and 4D thorax sCTs in terms of HU accuracy via proton radiography (PR).

Methods: PRs were simulated from a gantry angle of 0⁰ for pairs of sCTs and same-day repeat CTs (rCT) of 15 lung cancer patients who received proton therapy. 3D and 4D-sCTs were generated from thorax CBCTs, acquired with a 3D-acquisition protocol, using a deep convolutional neural network. 4D-rCTs, deformably registered to sCTs, were used as reference in terms of HU accuracy. Range error maps between PRs based on sCT and rCT were generated for three scenarios: (1) average 4D-rCT against 3D-sCT, (2) 4D-rCT against 4D-sCT in a 50% breathing phase, and (3) average 4D-rCT against average 4D-sCT. Mean (MRE) and standard deviations (SD) of the range errors maps were calculated for the whole anatomy, for regions in which lung tissue was present, and for all anatomy excluding lung tissue regions.

Results: MREs ranged between 0.0±1.0mm and -2.6±3.8mm, showing large variability across patients in terms of sCT HU accuracy (see figure 1). Comparable MREs and SDs were found in scenarios 1, 2 and 3, revealing similar HU accuracy in 3D-sCTs as compared to 4D-sCTs. Systematically positive MREs were observed in the whole anatomy. This effect was magnified in lung quantifications, suggesting the presence of a systematic shift in HU values in sCTs due to lung tissue inaccuracies.

Conclusion: PR is suitable as a tool to detect HU inaccuracies in thorax sCTs. In vivo PR acquisitions in thorax patients could serve as a quality control tool for thorax sCTs in a patient specific basis, advancing the implementation of sCTs into online adaptive proton therapy workflows.

Resources