A tool for semiautomatic evaluation of PET data for range verification in ion beam therapy

Introduction:

The positron emission tomography (PET) is a clinically proven method for verification of ion beam therapy. Due to fundamentally different physical processes leading to dose and activity the evaluation is based upon a comparison between the β+-emitter distribution measured during or after irradiation and a Monte-Carlo prediction from the treatment plan. A visual comparison slice by slice requires well trained personnel and is very time consuming. Furthermore, the reproducibility is low.
A software is presented that allows a semi-automatic and guided comparison of prediction an measurement to overcome the mentioned difficulties.

Material and Methods:

The range of the primary particles is the crucial parameter affecting the position of dose deposition and also the activity distribution. A direct extraction of the range from a β+-emitter distribution is not feasible, however, the range difference between two data sets can be determined. Therefore, a one dimensional range comparison algorithm was enhanced and extended to three dimensions. It generates a two dimensional matrix of range deviations, that represents the complete activity distribution.
A comprehensive software has been developed providing an intuitive graphical user interface to perform different evaluation methods. A statistical evaluation provides information about the overall agreement between measurement and prediction. System inherent fluctuations can be easily taken into account by using the statistic criterion. However, small deviations can be overlooked.
Hence the software provides different methods for analysis of local deviations. The matrix of range deviations is projected into the CT images to allow a fast localization of possible critical areas.
The gamma index is used to take stochastic fluctuations into account that are caused by the relatively low cont rates compared to the classic diagnostic usage of PET.
A common issue during fractionated ion beam therapy is the filling of cavities in the beam path due to physiological processes between two fractions. Such a change in the filling status can cause severe changes in the particle range and thereby a failure in dose deposition. Hence the filling of cavities is automatically evaluated, the result is presented graphically to the user.
For the development in-beam PET data captured at GSI Helmholtzzentrum für Schwerionenforschung, Darmstadt, Germany during patient treatment from 1997 to 2008 was used. All fractions from 6 patients were evaluated visually to check the performance of the range comparison algorithm and the feasibility of the different displaying possibilities of the software.
The feasibility of the statistical criterion has been tested on artificial range modifications. In simulated β+ -emitter distributions of 6 patients the range was shifted in the shape of a truncated quadratic pyramid with a maximum change of ± 4 mm, ± 6 mm and ± 10 mm water equivalent path length. These modified distributions have been compared to the measurement. By applying the statistical criterion a classification of the data sets was performed, i.e. the system had to decide whether a distribution was range modified.

Results:

From the classification results of the data sets the true positive and false positive rate was calculated to quantify the quality of the algorithm. For ±10 mm a true positive rate (TPR) of 90 % and a false positive rate (FPR) of 12 % and was reached, for ±6 mm a TPR of 82 % and a FPR of 26 % and for ± 4 mm a TPR of 72 % and an FPR of 26 %, respectively. This shows the abilities of the statistical criterion but also the limitations in the case of small deviations that make a local evaluation necessary. The comparison of the visual evaluation and the local results of the algorithm shows a good correlation. Differences in the result of the range comparison algorithm and the valuation of a human evaluator are rare.

Conclusion:

Despite system inherent difficulties like low count rates and physiological washout effects that blur the obtainable images, PET is a powerful tool for range verification in ion beam therapy. For a routine application a systematic and easy usable software tool is required. The developed tool combines a statistical approach to gain information about the overall agreement between measurement and prediction with local criterion. An intuitive graphical user interface allows also less trained personnel to evaluate the PET data.