Alan Priester1,2, Joshua Shubert1, Prateek Bhatnagar1, Jeremy Bong1, Shyam Natarajan1,2
1Avenda Health Inc.
2University of California, Los Angeles
Introduction: Partial gland ablation for treatment of prostate cancer (PCa) frequently relies upon placement of interstitial treatment applicators (ITAs)to generate ablation volumes(AVs). Planning ITA locations can be a complex and time-consuming process, since the optimal arrangement depends upon the ITA orientation, expected registration accuracy, AV size, and AV shape associated with the ablation modality. Herein we describe a fully automatic and modality-agnostic method for ablation site selection.
Methods: An algorithm was written in Python, requiring as inputs the ITA point of origin and 3D surfaces for the prostate, region of interest (ROI), margin intended for treatment, and ablation volume.
- The first AV was placed on the margin apical/basal edge closest to the ROI.
- AVs were placed sequentially to continuously encompass the margin’s lateral edge.
- AVs were placed sequentially to continuously encompass the margin’s medial edge.
- Additional AVs were placed via raster-scanning to encompass any central cavities.
- If a single AV was insufficient to encapsulate the margin along an ITA trajectory, additional “pullback” AVs were defined with even spacing (Fig C).
The algorithm was evaluated on 11 unilateral and 6 bilateral margins derived from 11 patients who had previously received focal treatment. The margin encapsulation of the AV aggregate was measured for 18 variations of AV size (transverse diameters=12,18, 26 mm) and geometry (spheres, ellipses, rounded cylinders i.e. “capsules”, cubes, prisms, and custom). Thus, a total of 18×17 = 306 instances were tested. Manually generated treatment plans using large custom AVs were available for the 11 unilateral margins; these were directly compared to automatically generated treatment plans using the same margins and AVs.
Results: Exemplary algorithm outputs are shown in Figure A-C. The algorithm completed after an average of 2.25 seconds (range 0.4-6.5) and placed an average of 11 AVs (range 1-41). The algorithm achieved mean margin coverage between 86% and 99% for each AV size/shape. Compared to manual planning, mean margin coverage was significantly improved from 87% to 94% (p = 0.005). Margin coverage was invariant to AV volume and margin volume. However, margin coverage increased for less rounded AV geometries (Fig D), due to the relative ease of continuously packing straight versus rounded contours.
Conclusion: This methodology could potentially be used to automate ablation site placement for a variety of treatment modalities, simplifying the planning process for focal treatment of prostate cancer.
