Can prompt-gamma-based verification detect anatomical changes in PT? First systematic clinical investigation

Introduction: Anatomical changes during proton therapy can cause severe dosimetric deviation. Treatment verification is thus highly desirable. Here, we present the first systematic evaluation of the sensitivity of a Prompt-Gamma-Imaging (PGI) based range verification system to detect anatomical changes in prostate-cancer treatments.

Materials and Methods: Spot-wise range deviations were monitored with a PGI slit camera during in total 16 fractions of hypo-fractionated Pencil-Beam-Scanning (PBS) prostate-cancer treatments (2 patients, 2 fields, each 1.5GyE). For all monitored fractions, in-room control-CT scans were acquired, serving as ground-truth reference for the identification and scoring of anatomical changes (strong/moderate/light). The sensitivity to detect these changes was determined for both, clinically measured and simulated PGI-data, respectively: For distal PBS spots, expected shifts, determined from line-dose profiles (planning-CT vs. control-CT), were manually compared with PGI-derived spot-wise shifts (Fig.1). Furthermore, a simple two-parametric model was established to classify each monitored field into scenarios of global, local and no-clinically-relevant anatomical changes.

Results: Overall 66% (84%) of the 64 detected anatomical changes were identified from measured (simulated) PGI-data (Fig.2a). All strong changes (14/64) were identified correctly. The first attempt for automated field-wise classification was able to correctly classify most global changes (9/11). However, differentiation between non-relevant from local changes seemed more difficult (4/6 and 7/14 fields classified correctly, respectively); but even ground-truth classification was often borderline in those cases (Fig.2b).

Conclusion: In the first systematic investigation of the sensitivity of clinical PGI-based treatment verification, its capability to detect strong anatomical changes has been clearly demonstrated. Moving towards automated interpretation of PGI-data, a simple two-parametric model already showed encouraging results.