Validation of non-rigid point-set registration methods using a porcine bladder pelvic phantom

Area of Science:

• Medical Physics
• Radiotherapy
• Image Registration

Background:

• Accurate dose accumulation in fractionated radiotherapy is crucial for treating deformable organs, such as the bladder.
• Current methods require improvement for robustness and efficiency.

Purpose of the Study:

• To evaluate the feasibility and accuracy of non-rigid (affine or deformable) point-set registration for bladder dose accumulation.
• To compare the performance of different registration algorithms on bladders of varying sizes and shapes.

Main Methods:

• A pelvic phantom with an ex vivo porcine bladder was used, with four different volume fillings (90-480 cc).
• Five MATLAB-implemented registration methods were compared: Coherent Point Drift (CPD), Gaussian Mixture Model, Shape Context, Thin-Plate Spline Robust Point Matching (TPS-RPM), and Finite Iterative Closest Point (ICP-finite).
• Evaluation metrics included registration runtime, Target Registration Error (TRE), Root-Mean-Square Error (RMS), and Hausdorff Distance (HD).

Main Results:

• All deformable algorithms yielded reasonable registration results.
• CPD achieved the best TRE (6.4 mm).
• TPS-RPM demonstrated the best mean RMS (1.4 mm) and HD (6.8 mm) values.
• ICP-finite was the fastest method, while TPS-RPM was the slowest.

Conclusions:

• Non-rigid point-set registration is feasible for accumulating dose in deformable bladders.
• CPD and TPS-RPM show promise for improving accuracy in radiotherapy dose accumulation.
• Algorithm selection depends on the balance between accuracy and computational time.