A method to optimize patient specific HU-SP calibration curve from proton range measurements

TL;DR

This paper introduces a fast, patient-specific method to optimize the Hounsfield units to stopping powers calibration curve in proton therapy using in-treatment range measurements, improving accuracy without extra dose.

Contribution

The proposed method uniquely uses in-treatment proton range data and a Taylor expansion approach to efficiently refine the calibration curve during therapy.

Findings

Reduced calibration uncertainty from 2.67% to 1.62%.

Lowered WEPL bias from 2.14% to 0.74%.

Effective with less than 10 treatment fractions.

Abstract

Proton radiotherapy promises accurate dose delivery to a tumor and minimal dose deposition to all other tissues. However, in practice the planned dose distribution may not conform to the actual one due to noisy data and different types of errors. One such error comes in a form of potentially inaccurate conversion of the Hounsfield units (HU) to stopping powers (SP) of protons. We propose a method of improving the CC based on a planning CT and proton range measurements acquired during treatment. The range data were simulated using a virtual CC and a planning CT. The range data were given two types of noise: range shift due to patient setup errors; and range noise due to measurement imprecision, including a misalignment of the range measuring device. The method consists of two parts. The first part involves a Taylor expansion of the water equivalent path length (WEPL) map in terms of the range shift caused by the difference between the planning and the virtual CC. The range shift is then solved for explicitly, leading to a polynomial function of the difference between the two CCs. The second part consists in minimizing a score function relating the range due to the virtual CC and the range due to the optimized CC. Tested on ten different CCs, our results show that, with range data collected over a few fractions (less than 10), the optimized CC leads to an overall reduction of the range difference. More precisely, on average, the uncertainty of the CC was reduced from 2.67% to 1.62%, while the average reduction of the WEPL bias was reduced from 2.14% to 0.74%. The advantage of our method over others is 1) its speed, and 2) the fact that the range data it necessitates are acquired during the treatment itself, and as such it does not burden the patient with additional dose.