Measurement-Based Pressure Model to Predict Organ Deformation during Histotripsy of the Liver
Abstract
Purpose
To develop a model of pressure distribution exerted by the histotripsy coupling water bath (with flexible membrane) to predict abdominal organ deformation for advanced histotripsy treatment planning.
Methods
The water bath (10-13L) was placed on n=8 healthy volunteers’ abdomens. Relative pressure was measured using a force sensor (15mm diameter) on a 44x32cm grid (2.0x1.4cm resolution) with the total force separately estimated by measuring the weight on a scale as a function of water volume. A predictive data-driven model of pressure distribution was created using a linear combination of 1) local water height, 2) distance to the rib edge, 3) local topography (point height relative to surrounding neighborhood; radii 4 and 8cm), 4) a binary indicator for points within 5 mm of the contact-interface edge, and 5) a binary indicator for anterior surface points overlying ribs. Alternative pressure models were investigated: uniform and hydrostatic (P=density*gravitational acceleration*height). The three pressure models were evaluated for accuracy by comparing predictions to measurements and computing RMSE. For the data‑driven model, evaluation was performed using leave-one-subject-out cross-validation. To evaluate the effect of pressure models on deformation, finite element (FE) meshes of n=8 humans were generated from CT-derived segmentations (TotalSegmentator dataset) and deformation predicted using each pressure model, assuming 12L water. Deformation differences between the data-driven model and alternatives were evaluated throughout the liver (1.5mm resolution).
Results
RMSEs were 0.41±0.11kPa, 0.52±0.12kPa, and 1.02±0.21kPa for data-driven, uniform, and hydrostatic models, with the data‑driven model significantly outperforming alternatives (p=0.023;1.3×10⁻⁵). Mean Euclidean differences between deformation induced by data-driven models and alternatives were statistically significant: 2.1±0.5mm for uniform (p=6×10⁻6) and 4.0±0.8mm for hydrostatic (p=2×10⁻6). Differences ranged up to 13.6mm.
Conclusion
The data-driven pressure model was significantly more accurate and resulted in significantly different FE-based liver deformation compared to alternative pressure models. FE-predicted deformation using this model will be evaluated for accuracy in future studies.