Machine learning-based patient specific prompt-gamma dose monitoring in proton therapy

Abstract

Online dose monitoring in proton therapy is currently being investigated with prompt-gamma (PG) devices. PG emission was shown to be correlated with dose deposition. This relationship is mostly unknown under real conditions. We propose a machine learning approach based on simulations to create optimized treatment-specific classifiers that detect discrepancies between planned and delivered dose. Simulations were performed with the Monte-Carlo platform Gate/Geant4 for a spot-scanning proton therapy treatment and a PG camera prototype currently under investigation. The method first builds a learning set of perturbed situations corresponding to a range of patient translation. This set is then used to train a combined classifier using distal falloff and registered correlation measures. Classifier performances were evaluated using receiver operating characteristic curves and maximum associated specificity and sensitivity. A leave-one-out study showed that it is possible to detect discrepancies of 5 mm with specificity and sensitivity of 85% whereas using only distal falloff decreases the sensitivity down to 77% on the same data set. The proposed method could help to evaluate performance and to optimize the design of PG monitoring devices. It is generic: other learning sets of deviations, other measures and other types of classifiers could be studied to potentially reach better performance. At the moment, the main limitation lies in the computation time needed to perform the simulations.

Technical Note: Machine learning approaches for range and dose verification in proton therapy using proton‐induced positron emitters

“…This is consistent with the previous studies for the range verification reporting an achievable accuracy of ~1–2 mm. A similar machine learning‐based study used the direct threshold method based on the measurement of prompt gamma profiles, obtained an accuracy of ~5 mm …”

“…[1][2][3][4] The rational is that the spatial distribution of secondary signals correlates with both dose distribution and proton range. A number of methods have been explored, including positron radioisotope, 5-10 prompt gamma (PG), [11][12][13][14][15] acoustic wave, 16,17 and Cherenkov photons. 18,19 As a promising tool, Positron Emission Tomography (PET) was extensively studied where the distribution of positron emitters (e.g., 11 C, 15 O) produced can be used to obtain dose distribution.…”

The physics of proton therapy

“…In cases where uncertainties are presently too large, what roles will prospective imaging play, including proton CT (Schulte et al , 2004) and megavoltage photon CT (Langen et al , 2005; Newhauser et al , 2008a; De Marzi et al , 2013)? Similarly, what role will real-time or post-treatment imaging play, including prompt gamma imaging (Peterson et al , 2010; Gueth et al , 2013), positron emission tomography (Parodi et al , 2007; Cho et al , 2013; Min et al , 2013), proton radiography (Schneider and Pedroni, 1995), and magnetic resonance imaging (Krejcarek et al , 2007)?In the future, what out-of-field dose algorithms should be developed for treatment planning systems used for research or routine clinical practice? Techniques for the calculation and measurement of therapeutic dose are reasonably well established (with a few exceptions mentioned below).…”

Detecting prompt gamma emission during proton therapy: the effects of detector size and distance from the patient

“…Recent studies by several research groups have shown a strong correlation between the region of PG emission and dose deposition by the treatment beam, including both measurement and Monte Carlo (MC) studies in simple phantoms (Bom et al , 2012; Kim et al , 2009; Kormoll et al , 2011; Le Foulher et al , 2010; Min et al , 2006; Roellinghoff et al , 2011; Verburg et al , 2013), as well as MC studies in CT based patient geometry (Gueth et al , 2013; Moteabbed et al , 2011). The findings of these studies have led to an increased interest in PG emission as a means of measuring the in vivo range of proton beams and verifying treatment delivery, thus reducing the magnitude of the uncertainties associated with proton therapy.…”