KEY POINTS
- This retrospective in-silico study included 25 patients with abdominal-pelvic lymph-node oligometastases, all originally treated with five-fraction SBRT under the STEAL protocol. Investigators simulated online-adaptive treatment using a 9-field IMRT prescription of 5 × 9 Gy.
- Five accumulation approaches were compared: conventional DVH summation, rigid registration, image-based deformable registration, hybrid deformable registration and Thin Plate Spline–Robust Point Matching (TPS-RPM). Across 34 gastrointestinal-organ segments, 2,360 fraction-to-fraction registrations were evaluated.
- Registration quality improved dramatically with algorithmic complexity. TPS-RPM achieved median Dice 0.975, HD99 1.35 mm and mean surface distance 0.254 mm, compared with Dice 0.615 and mean surface distance 4.27 mm for rigid registration.
- Better geometry changed the estimated accumulated bowel dose, but modestly. TPS-RPM produced median gastrointestinal D0.5cc 25.5 Gy versus 26.9 Gy with DVH summation, and V20Gy3 5.4 versus 5.9 cc. The physical D0.5cc estimate was approximately 5.5% lower with TPS-RPM.
- Crucially, those differences did not change the fraction-reduction strategy. Both TPS-RPM and simple DVH summation identified exactly the same 9 of 25 patients (36%) for treatment shortening: four could finish in two fractions, four in three and one in four, reducing total fractions across the cohort by 16.8%.
- Among the remaining 16 patients, dose-escalating adaptation increased PTV V100 from 83.4% to 91.4% with DVH summation and 91.6% with TPS-RPM, with no significant difference between approaches. PTV D95 increased from 55.4 Gy10 to 73.7 versus 78.5 Gy10, respectively; this was the only target metric significantly favoring TPS-RPM.
- No difference in fraction-reduction decisions and minimal differences in target escalation led the authors to question whether the added implementation and validation burden of advanced deformable accumulation is justified for this particular adaptive workflow. The work remains an in-silico study and does not establish equivalence for other disease sites or accumulation-sensitive endpoints.
CLINICAL TAKEAWAY
The most geometrically accurate solution is not automatically the most useful clinical solution. Here, sophisticated deformable accumulation produced cleaner anatomy and slightly lower estimated bowel doses, yet simple DVH summation led to essentially the same adaptive treatment decisions. That makes a strong argument for evaluating algorithms by decision impact, not geometric metrics alone.