Electron-density assessment using dual-energy CT: accuracy and robustness

Purpose/Objective:

Current treatment planning for essentially every external radiation therapy (photons, electrons, protons, heavier ions) is not able to account for patient-specific tissue variability or non-tissue materials (e.g. implants, contrast agent) which can lead to considerable differences in dose distributions (figure 1). This is due to the conversion of CT numbers to electron density or stopping power using a heuristic Hounsfield look-up table. In contrast, dual-energy CT (DECT) allows for a patient-specific determination of electron density – the only (most important) parameter influencing photon (ion) dose distributions. Among the many algorithms proposed for this purpose, a trend towards increased complexity is observed, which is not necessarily accompanied by increased accuracy and might at the same time militate against clinical implementation. Here, we therefore investigated the performance of a seemingly simple linear-superposition method (Saito, 2012, Hünemohr et al., 2014).

Material/methods:

Key feature of the studied approach is a parameterization of the electron density, given by 'alpha blending” of the two DECT images. The blending parameter can be obtained by empirical calibration using a set of bone tissue surrogates and a linear relationship between relative photon absorption cross sections of the higher and lower voltage spectrum. First, this linear relation was analyzed to quantify the purely methodological uncertainty (i.e. with ideal CT numbers as input), based on calculated spectral-weighted cross sections from the NIST XCOM database for tabulated reference tissues (Woodard and White, 1986). A clear separation from CT-related sources of uncertainty (e.g. noise, beam hardening) is hereby crucial for a conclusive assessment of accuracy. Secondly, we tested the proposed calibration method on published DECT measurements of typical tissue-surrogate phantoms and evaluated its uncertainty.

Results:

The methodological uncertainty of electron-density assessment for the alpha-blending method was found to be below 0.15% for arbitrary mixtures of human tissue. In the case of small abundance of high-Z elements, electron-density results are positively biased, e.g. 0.5% for thyroid containing 0.1% iodine (Z=53) by mass, which is due to the K edge of the photoelectric effect. The calibration parameters obtained from various published data sets, showed very little variation in spite of diverse experimental setups and CT protocols used. The calibration uncertainty was found to be negligible for soft tissue while it was dominated by beam hardening effects for bony tissue.

Conclusion:

The alpha-blending approach for electron-density determination shows universal applicability to any mixture of human tissue with a very small methodological uncertainty (< 0.15%); and a robust and bias-free calibration method, which is straightforward to implement. We conclude that further refinement of algorithms for DECT-based electron-density assessment is not advisable.